Sinkron : jurnal dan penelitian teknik informatika

Enhancing Entity Extraction in E-Government Complaint Data using LDA-Assisted NER

Thu, 02 Oct 2025 00:00:00 +0000

With the rapid development of information technology, governments are increasingly challenged to provide digital channels that enhance public participation in governance. LaporGub, an official platform managed by the Central Java Provincial Government, accommodates citizens' aspirations and complaints, but faces challenges in processing large amounts of unstructured text. Manual analysis is time-consuming and error-prone, resulting in delayed responses and decreased service quality. Conventional Named Entity Recognition (NER) models struggle to handle informal Indonesian-language text, while transformer-based approaches require substantial computing resources that are not widely available in local government environments. Therefore, this study aims to develop a lightweight NER approach by integrating Latent Dirichlet Allocation (LDA) as a semantic pre-annotation tool to improve the accuracy of entity extraction in Indonesian e-government complaint data. To achieve this goal, a dataset of 53,858 complaint reports from the LaporGub platform (2022–2025) was processed using LDA topic modeling (k=10) to provide semantic context during annotation. Next, the enriched dataset was used to train a spaCy-based NER model targeting three entity types: LOCATION, ORGANIZATION, and PERSON, with a training-validation-test split ratio of 70:15:15 using stratified sampling. The evaluation showed that the proposed NER+LDA model achieved a precision of 90.03%, a recall of 81.86%, and an F1-score of 85.75%, representing improvements of +5.78, +2.55, and +4.04, respectively, compared to the baseline NER model (F1-score: 81.71%). Furthermore, the most significant improvements occurred in the detection of ORGANIZATION and PERSON entities. These findings confirm that the integration of LDA as a pre-annotation strategy effectively improves NER performance on informal complaint texts in Indonesia, thus offering a practical and resource-efficient alternative to transformer-based methods for e-government applications.

Integrating K-Means Clustering and Apriori for Data Mining-Based Digital Marketing Strategy For Increasing UMKM: Study Case Stabat City

Adek Maulidya Maulidya, Selfira, Gomgom Sidabutar, Reyva Ryo Al Hafiz — Fri, 17 Oct 2025 00:00:00 +0000

Micro, Small, and Medium Enterprises (MSMEs) or UMKM in Bahasa are play a crucial role in regional economic development, yet they often face challenges in designing effective marketing strategies due to limited access to advanced analytical tools. Digital marketing supported by data mining offers a solution to this problem by enabling more precise customer segmentation and product bundling recommendations. This study aims to integrate K-Means clustering and Apriori association rule mining to develop data-driven marketing strategies for MSMEs in Stabat City, Indonesia, with a specific focus on rice sales data. A dataset consisting of 1,000 rice sales transactions was processed through a multi-stage methodology, including data preprocessing, clustering, and association rule generation. The Elbow and Silhouette methods suggested an optimal cluster number of k = 3, resulting in three distinct customer groups: (1) loyal high-value buyers, (2) price-sensitive buyers, and (3) premium-oriented buyers. Descriptive statistics highlighted differences in average transaction values, purchase frequency, and brand preferences across clusters. Apriori analysis produced the top ten significant association rules, such as {Medium Rice} → {Pandan Wangi Rice} with support = 0.14, confidence = 0.68, and lift = 1.23. Promotional simulations showed that generic discount campaigns could increase sales by approximately 3.0%, whereas targeted bundling strategies yielded smaller short-term gains (+1.53%) but offered stronger long-term potential, particularly for premium-oriented clusters. These findings are consistent with prior international studies, where customer segmentation combined with market basket analysis has proven effective for enhancing digital marketing outcomes. The study concludes that integrating clustering and association rules can provide MSMEs with actionable insights to optimize promotional strategies and improve competitiveness. However, limitations remain, including the relatively small dataset, reliance on manual parameter selection, and simplified modeling assumptions. Future research should expand to multi-sector datasets and explore advanced algorithms to validate and extend these findings.

Hybrid GA–MILP Model for Community Building Retrofit Planning Towards Carbon Neutrality

Chairini Aisyah, Adhita Nugraha Mestika — Wed, 15 Oct 2025 00:00:00 +0000

Retrofitting community buildings is a key pathway toward carbon neutrality, yet most existing retrofit planning models lack adaptability to the diverse urban contexts of the Global South, where building typologies are heterogeneous and resources limited. Addressing this gap requires approaches that are both computationally efficient and context-sensitive. This study introduces a hybrid optimization framework that integrates Genetic Algorithm (GA) and Mixed-Integer Linear Programming (MILP) to tackle the multidimensional multiple-choice knapsack problem inherent in retrofit planning. The GA explores high-level system configurations, while MILP ensures precise component-level selection under budget and technical constraints. Compared to conventional single-method approaches, the hybrid GA–MILP achieves near-optimal emission reduction with reduced computation time and greater feasibility, offering a balanced trade-off between performance and scalability. Importantly, the framework demonstrates that medium-cost retrofit strategies provide the most cost-effective path to scalable carbon savings, making it highly relevant for resource-constrained urban environments. By situating retrofit planning within the realities of the Global South, this study advances methodological innovation and provides a robust decision-support tool aligned with sustainable development goals for inclusive and low-carbon urban futures.

Comparative Analysis of DNA Sequence Alignment Algorithms in SARS-CoV-2

Edi, Robet, Nurhayati Harahap — Wed, 22 Oct 2025 00:00:00 +0000

Sequence alignment is fundamental in bioinformatics, with Smith-Waterman (local) and Needleman-Wunsch (global) algorithms widely applied. However, comparative analyses on highly similar viral genomes such as SARS-CoV-2 remain scarce. This study systematically evaluated both algorithms using the first 5,000 nucleotides of two SARS-CoV-2 genomes (29,903 and 29,684 nt) under four parameter configurations: standard, low gap penalty, high gap penalty, and high match reward. Performance was assessed through alignment score, sequence identity, gap distribution, execution time, and parameter sensitivity. Both algorithms produced identical sequence identity (97.80%), with 4,943 matches out of 5,054 positions. Smith-Waterman consistently yielded higher alignment scores (12.6-112 points advantage), while Needleman-Wunsch was substantially faster (0.7752 vs 3.9014 s), showing 5.03 times greater computational efficiency. These findings indicate that both methods are reliable for highly similar viral sequences, with a trade-off between scoring precision and computational speed. This study provides the first parameter-sensitive comparison for full SARS-CoV02 genomes, emphasizing how parameter tuning can influence performance outcomes. A key limitation is that the analysis was restricted to the first 5,000 nucleotides, which may not capture variability across the complete genome.

Research and Analysis of Exchange Sort Algorithm in Data Structure

Rakhmat Purnomo, Tri Dharma Putra — Thu, 02 Oct 2025 00:00:00 +0000

Exchange sort is different from bubble sort. Exchange sort compares an element with other elements in the array, and swaps elements if necessary. So there is an element that is always the center element (pivot). Here is its theoretical description: Comparison, the algorithm compares each element with its adjacent element. Then continue until all elements are compared. Swap: If the elements are in the wrong order (for example, in ascending order, if the left element is greater than the right), they are swapped. This swapping continues until all match numbers are swapped. Iteration, this process of comparing and swapping, is repeated for each pair of adjacent elements in the array. Looping, this process is repeated a number of times (traversing) the array until no more swapping is required, indicating that the array is sorted. It is concluded that for the six numbers in these three case studies, the iterations needed are 5 iterations each. The swaps counts needed are 7, for case study 1. The swap counts needed are 12 for case study 2 and the swap counts are 8 for case study 3. In this research and analysis, the order, all of them is descending, although it can be made ascending. In modern days, exchange sort plays a very important role in terms of sorting algorithms. This paper is only research and analysis. For novelty, the analysis is given with a clear step-by-step procedure of the algorithm.

A Hybrid Three-Term Conjugate Gradient Algorithm for Solving Unconstrained Optimization Problems

Radhwan Basem Thanoon — Sat, 18 Oct 2025 00:00:00 +0000

In this paper, we introduce a novel hybrid three-term conjugate gradient algorithm referred to as THREER, designed to address unconstrained optimization problems. The proposed approach integrates the -parameter introduced by Al-Neami with an additional third component derived from a rate-based vector , resulting in a search direction that preserves and enhances key characteristics of traditional conjugate gradient methods. A rigorous theoretical investigation establishes that the algorithm satisfies the sufficient descent condition regardless of the line search technique employed. Furthermore, the global convergence of the method is guaranteed under commonly accepted assumptions. Extensive numerical experiments conducted on large-scale benchmark problems reveal that THREER achieves superior performance when compared with several classical algorithms, particularly in terms of iteration count and function evaluations. These results highlight the algorithm’s robustness, efficiency, and potential for solving high-dimensional optimization tasks.

Implementation of an Integrated Cloud-Based Electronic Medical Record System at Community Health Center

Raja Nasrul Fu’ad, Sugeng Riyadi, Deny Pratama, Ramadhan Wicaksono — Fri, 31 Oct 2025 00:00:00 +0000

Klambir Lima is a village located in Hamparan Perak District, Deli Serdang Regency, North Sumatra Province. It is a small village with a dense population but has only one government community health center (puskesmas). This results in suboptimal patient services. Furthermore, there is no existing application for recording patient data or medical records that could assist in patient data management. Patient medical records are a crucial feature in healthcare services. They are useful for recording or storing a patient's health or illness history, which enables accurate treatment or medication tailored to the patient's needs. Therefore, the Klambir Lima Health Center requires an electronic medical records application based on Cloud Computing. Data will be stored in cloud storage, aiming to minimize damage or loss of data, which is a vital asset. In this research, the author developed the application using the R&D (Research and Development) method. The role and utility of the research were well established to ensure better implementation of the application. The research objective is to create a system capable of recording electronic-based medical records via cloud computing media. This will enable the Klambir Lima Health Center to improve its healthcare services for both BPJS (national health insurance) and non-BPJS patients

Addressing Class Imbalance in Stunting Classification Using SMOTE Enhanced Random Forest

Thu, 09 Oct 2025 00:00:00 +0000

Stunting is a chronic nutritional problem that poses serious long-term effects on children’s health, including impaired physical growth, delayed cognitive development, and reduced productivity in adulthood. Early and accurate detection of stunting is therefore essential to support effective public health interventions and targeted policy implementation. However, one of the central challenges in developing machine learning models for this purpose is the presence of class imbalance in health-related datasets. Such imbalance frequently leads to biased classifiers that perform well on majority classes but fail to identify minority categories, reducing the overall reliability of the system. To overcome this issue, the present study utilized the Synthetic Minority Oversampling Technique (SMOTE) to balance the distribution of classes in a dataset containing 110,000 records. A Random Forest algorithm was then employed as the base classifier, with hyperparameter optimization carried out using the Optuna framework to ensure robustness and generalizability. The experimental results demonstrate that the combined application of SMOTE and Optuna significantly improved classification performance, producing the highest Macro Area Under the Curve (AUC) of 0.9972. This outstanding score indicates the model’s superior ability to distinguish nutritional status categories across both majority and minority classes. The study concludes that addressing data imbalance through oversampling is a fundamental methodological step in constructing fair and effective machine learning systems for stunting detection, ultimately contributing to improved health outcomes and evidence-based policy design.

Integrating Bayesian Optimization into Ensemble Logistic Regression for Explainable AI-Based Customer Behavior Analysis

Jeffry, Azminuddin I. S. Azis, Elisabeth Tri Juliana Kandakon — Thu, 02 Oct 2025 00:00:00 +0000

Understanding customer behavior is a strategic factor in business decision-making, particularly within the automotive sector, where competition is intense and product variety is diverse. While previous studies often rely on limited demographic variables, such as age and gender, this research advances the field by integrating ensemble logistic regression with Bayesian Optimization for hyperparameter tuning and SHAP-based interpretability. The proposed model incorporates additional features beyond demographics, including vehicle category, product type, vehicle year, dealer branch, and transaction source, to enhance predictive accuracy. The methodology involves data preprocessing through encoding and cleaning, class balancing using SMOTE combined with undersampling, and stratified train-test splitting (80:20). Baseline Logistic Regression achieved an accuracy of 80%, ROC AUC of 0.89, precision of 0.47/0.96, recall of 0.84/0.79, and F1-scores of 0.59/0.89. By applying ensemble logistic regression with Bayesian Optimization, performance improved to 84% accuracy, ROC AUC of 0.92, precision of 0.51/0.98, recall of 0.83/0.84, and F1-scores of 0.63/0.92. SHAP analysis confirmed that the additional features significantly contribute to prediction outcomes. The novelty of this study lies in combining Ensemble Logistic Regression with Bayesian Optimization and SHAP explainability in the automotive domain, offering not only improved accuracy but also interpretability and fairness for business decision-making, providing actionable insights for targeted marketing strategies and product management. Future studies may incorporate broader behavioral and transactional variables to capture more nuanced customer decision patterns..

Lightweight Deep Learning Models for Facial Expression Recognition in Inclusive Education

Miftahul Ilmi, Doni Syofiawan — Tue, 28 Oct 2025 00:00:00 +0000

Facial expression recognition is an essential component in the development of artificial intelligence-based learning systems, particularly in the context of inclusive education that involves students with special needs. This study aims to evaluate the performance of several lightweight deep learning architectures in detecting facial expressions with high accuracy while maintaining computational efficiency. Facial image data were obtained from both public datasets and newly collected samples, which were preprocessed through face cropping, normalization, and data augmentation. The dataset was split into 70% training, 15% validation, and 15% testing. Four lightweight deep learning architectures: MobileNetV2, MobileNetV3 (Small and Large), and EfficientNetB0, were employed as the primary models using transfer learning and fine-tuning approaches. Evaluation was conducted using accuracy, loss, precision, recall, and F1-score metrics, complemented by visualization through confusion matrices. The results indicate that MobileNetV2 achieved the best performance with a test accuracy of 92%, precision of 93%, recall of 91%, and F1-score of 92%, while maintaining a relatively lightweight parameter size of 2.26 million. EfficientNetB0 ranked second with 83% accuracy, followed by MobileNetV3-Large (77%), whereas MobileNetV3-Small demonstrated the lowest performance (45%). Confusion matrix analysis revealed recurring misclassification patterns for certain expressions, such as Happy often misclassified as Sad, and Neutral overlapping with Angry. This study confirms that MobileNetV2 is the most optimal architecture for implementing facial expression recognition systems in inclusive education environments, as it balances high accuracy with computational efficiency. These findings provide a solid foundation for developing intelligent applications that support adaptive interaction in the learning process..

Association Rule Mining across Multiple Domains: Systematic Literature Review

Dayini Syahirah, Priati, Okky Pratama Martadireja — Thu, 02 Oct 2025 00:00:00 +0000

This Systematic Literature Review (SLR) synthesizes 50 studies published between 2020 and 2025 that applied Association Rule Mining (ARM) across multiple domains, using the PRISMA 2020 framework. The review examines application areas, algorithm choices, implementation tools, parameter settings, and emerging trends. Results indicate that transportation and market analysis are the most prominent domains, followed by healthcare, manufacturing, and governance, with smaller contributions from tourism, agriculture, energy, and the environment. Apriori remains the most widely used algorithm due to its simplicity, FP-Growth is preferred for efficiency, and hybrid or modified approaches are adopted to address scalability issues. Python dominates as the primary implementation tool, alongside RapidMiner and R-Studio, with parameter thresholds generally adapted to dataset size and domain-specific needs. The novelty of this review lies in providing a cross-domain synthesis of ARM, filling the gap left by prior reviews that were limited to specific fields or algorithms. This broader perspective reveals temporal trends and recurring challenges, particularly scalability and interpretability, while identifying opportunities such as integration with deep learning, real-time ARM, and cross-domain adaptation. By offering a structured overview of developments in ARM, this study contributes both conceptual insights and practical guidance, serving as a reference for optimizing applications and informing future research directions.

Integrating Blockchain with Neural Networks for Forest Fire Classification

Hernan Yudistira, Djarot Hindarto, Asrul Sani — Sun, 02 Nov 2025 00:00:00 +0000

Forest fires represent one of the most severe environmental disasters, causing extensive ecological, social, and economic damage—particularly in tropical nations like Indonesia. This research introduces a hybrid framework that combines Blockchain and Neural Network technologies to classify forest fire images. The goal is not only to enhance detection precision but also to guarantee the integrity and security of experimental data. Two deep learning architectures, ResNet-50 and VGG-16, were implemented and evaluated to compare their effectiveness in differentiating fire from non-fire imagery. The dataset merges locally collected images from the Puncak area of Bogor, Indonesia, with the public FIRE dataset from Kaggle, thereby increasing model generalization. Model training utilized a transfer learning strategy, and its performance was assessed through four key indicators: accuracy, precision, recall, and F1-score. The findings demonstrate that VGG-16 achieved the most reliable outcomes, obtaining an accuracy of 0.91 and an F1-score of 0.87, outperforming ResNet-50, which reached 0.82 accuracy. All experimental data, including training and inference outputs, were stored using the InterPlanetary File System (IPFS), while each file’s Content Identifier (CID) and metadata were recorded in a blockchain-based smart contract to ensure transparency, verifiability, and data permanence. The study concludes that integrating blockchain with deep learning establishes a trustworthy and tamper-resistant framework for forest fire image classification. Future research may explore lighter CNN models and the fusion of IoT sensor data to enable adaptive and real-time fire detection.

Attention Augmented Deep Learning Model for Enhanced Feature Extraction in Cacao Disease Recognition

Robet, Johanes Terang Kita Perangin Angin, Tarq Hilmar Siregar — Thu, 02 Oct 2025 00:00:00 +0000

Accurate cacao disease recognition is critical for safeguarding yields and reducing losses. Prior cacao studies primarily rely on handcrafted descriptors (eg, Color Histogram, LBP, GLCM) or standard CNN/transfer-learning pipelines, often limited to ≤ 3 classes and a single plant organ; explicit channel-spatial attention and comprehensive multiclass evaluation remain uncommon. To the best of our knowledge, no prior work integrates Squeeze-and-Excitation (SE) and the Convolutional Block Attention Module (CBAM) on a ResNeXt50 backbone for six-class cacao disease classification, accompanied by a standardized ablation study and t-SNE-based interpretability. We propose a six-class classifier (five diseases + healthy) built on ResNeXt-50 enhanced with SE (channel recalibration) and CBAM (channel-spatial emphasis) to highlight lesion-relevant cues. The dataset comprises labeled leaf and pod images from public sources collected under field-like conditions; preprocessing includes resizing to 224x224, normalization, and augmentation (flips, small rotations, color jitter, random resized crops). Trained with Adam and early stopping, ResNeXt50+SE+CBAM attains 97% test accuracy and 0.97 macro-F1, surpassing a ResNeXt50 baseline of 94% and 0.95 and SE-only/CBAM-only variants. Confusion matrix and t-SNE analyses show fewer mix-ups among visual classes and clearer separability, while the ablation validates complementary benefits of SE and CBAM. On a desktop-hosted, web-based setup, batch-1 inference at 224x224 is 7.46 ms/image (134 FPS), demonstrating real-time capability. The findings support deployment as browser-based decision-support tools for farmers and integration into continuous field-monitoring systems.

MCDM-Based Blockchain and Artificial Intelligence Integration for Earthquake Risk Recommendation System

Aditya Widianto, Ratih Titi Komala Sari, Djarot Hindarto , Asrul Sani — Sun, 02 Nov 2025 00:00:00 +0000

Indonesia is one of the countries with the highest earthquake vulnerability in the world because it is located at the meeting point of three major tectonic plates, namely Eurasia, Indo-Australia, and Pacific. The high risk of disaster requires a system that is capable of analyzing, predicting, and recommending earthquake-prone areas accurately, efficiently, and safely. This study aims to develop an earthquake risk recommendation system based on the integration of Artificial Intelligence (AI), Multi-Criteria Decision Making (MCDM), and Ethereum Blockchain. Earthquake data was obtained from Google Earth Engine (GEE) and geospatial data from the Geospatial Information Agency (BIG) and BMKG. The data is processed using AI algorithms for predictive analysis, then the MCDM methods of TOPSIS, and ELECTRE are applied to determine the priority of earthquake-prone areas based on a combination of seismic parameters, population density, infrastructure vulnerability, and distance to active faults. The analysis results are stored in a decentralized manner using the Ethereum Blockchain through smart contracts to ensure data integrity, security, and transparency. The research results show that the integration of AI–MCDM is capable of providing earthquake risk recommendations with high accuracy, while the application of blockchain ensures that the results cannot be manipulated. This system is expected to become a decision-making tool for disaster management agencies such as BMKG and BNPB in data-based earthquake risk mitigation.

Optimizing Supplier Selection Through Hybrid BWM and AHP Integration

Afrizal Rhamadan Siregar, Hendry Hendry — Thu, 02 Oct 2025 00:00:00 +0000

This study proposes a hybrid decision-making model that integrates the Best-Worst Method (BWM) with the Analytic Hierarchy Process (AHP) to optimize supplier selection. The primary objective is to address limitations in traditional Multi-Criteria Decision-Making (MCDM) methods, such as inconsistency, subjectivity, and cognitive overload when handling complex criteria. The proposed model leverages AHP's hierarchical structuring and BWM’s efficiency in reducing comparison load, aiming for a more accurate and consistent evaluation framework. The research design involves developing a hybrid AHP-BWM model and applying it to a dataset from the Vietnamese Textile and Apparel (T&A) sector. The methodology includes two stages: determining the weight of each criterion using a Hesitant-AHP approach, followed by evaluating supplier alternatives with BWM. The performance of the model is assessed using classification metrics, namely accuracy, precision, recall, and F1-score. The results show that the proposed model outperforms conventional methods such as TOPSIS, ELECTRE, VIKOR, and SWARA. It achieves an accuracy of 92%, precision of 87%, recall of 86%, and an F1-score of 86%. These outcomes confirm the model’s superior ability to consistently classify supplier suitability. Furthermore, the model identifies Quality Assurance as the most critical criterion, followed by Assistance, Capacity, Charge, and Shipment. In conclusion, the hybrid AHP-BWM model offers a robust, scalable, and data-driven approach for supplier selection. Its strength lies in balancing systematic evaluation with reduced cognitive effort, making it suitable for complex real-world decision-making environments. Future research may explore its application in other domains and enhance its scalability for larger datasets.

Implementation of a Hybrid Cryptosystem Using ChaCha20 and ECC for Image Encryption in an Android Application

Thu, 02 Oct 2025 00:00:00 +0000

This study aims to develop an Android application capable of securely encrypting and decrypting images using a hybrid cryptographic method. The system combines the ChaCha20 algorithm as symmetric cryptography to encrypt image files, and Elliptic Curve Cryptography (ECC) as asymmetric cryptography to encrypt the ChaCha20 key. The key used is temporary (ephemeral), ensuring that only the intended recipient who possesses the appropriate ECC private key can decrypt the file. The application was developed using the Kotlin programming language in Android Studio, with a PHP-based backend and MySQL database. Testing was conducted using the black-box method and involved 15 beta testers to evaluate functionality, security, and usability aspects. The results show that all features of the application run properly, and the encryption and decryption processes can be performed efficiently and securely. Beta testers gave an average rating of 4.6 out of 5 and stated that the application is easy to use and provides sufficient protection for personal data. Therefore, the developed application successfully meets the objectives of the study and offers an alternative solution for securing image file transfers between users via Android devices.

Multiple Linier Regression Analysis Effects of Education Media on Student Experience and Satisfaction

Gunturari Wibowo, Bambang Purnomosidi D P, Widyastuti Andriyani — Fri, 03 Oct 2025 00:00:00 +0000

The purpose of this study is to explore the effect of quiz- and game-based learning media on student learning experiences and satisfaction. This study models the relationship between media, motivation, experience, and learning satisfaction differently from previous studies with single variables. The research method uses a multiple linear regression approach with data collected through a Likert scale questionnaire. The research subjects consisted of 107 tenth-grade students majoring in Network and Computer Engineering at SMK Bhinneka Karya Simo. Data processing was carried out using Python and Microsoft Excel software for data analysis. The results showed that the variables of media, motivation, and response had a significant effect on learning experience (p < 0.05). The regression equations obtained were: experience (Y2) = 0.5918 + 0.2594X1 + 0.2840Y1, while satisfaction (Y3) = 0.5918 + 0.2594X1 + 0.2840Y1 + 0.3239Y2. In conclusion, learning experience is mainly influenced by media and motivation, while learning satisfaction is influenced by media, motivation, and the experience itself. These findings confirm that game-based learning strategies can create more meaningful learning experiences and encourage increased student satisfaction, which can be used as a basis for improving the quality of learning in the classroom.

Explainable Machine Learning for Poverty Prediction in Central Java Regencies and Cities

Wahyu Fhaldian, Amiq Fahmi — Sat, 04 Oct 2025 00:00:00 +0000

Poverty remains a multidimensional challenge in Central Java, necessitating robust data-driven approaches to identify its socioeconomic determinants. This study applied six machine learning models, specifically Extreme Gradient Boosting (XGBoost), Random Forest, CatBoost, LightGBM, Elastic Net Regression, and a Stacking ensemble using district-level data from Statistics Indonesia covering demographics, education, labor, infrastructure, and household welfare. Model evaluation combined an 80:20 hold-out split, 10-fold cross-validation, and noise perturbation tests. Results show that XGBoost achieved the best individual performance (MAE = 2,180.01; RMSE = 3,512.07; R² = 0.931), while the Stacking ensemble surpassed all single learners (MAE = 2,640.99; RMSE = 3,202.79; R² = 0.942). Interpretability was ensured through SHAP (Shapley Additive Explanations), Partial Dependence Plots (PDP), and Accumulated Local Effects (ALE), consistently identifying Number of Households, Per Capita Expenditure, and Uninhabitable Houses as the most influential predictors. Counterfactual simulations indicated that increasing per capita expenditure by 10% could reduce the poverty index by 9.9%, while reducing household size by 10% lowered it by 11.3%. Robustness checks revealed Brebes as an influential district shaping model stability. Overall, the findings demonstrate that boosting and stacking ensembles, when combined with explainable AI tools, not only enhance predictive accuracy but also provide transparent, policy-relevant evidence to strengthen poverty alleviation programs in Central Java. This study contributes both methodological advances in explainable machine learning and practical insights for targeted poverty reduction strategies.

A Disaster-Aware Traffic Assignment Model: Comparative Evaluation of Frank-Wolfe and Simulated Annealing Algorithms

Suranto, Afrizal Rhamadan Siregar — Fri, 10 Oct 2025 00:00:00 +0000

Traffic assignment under disaster-induced disruptions poses unique challenges, as traditional models often overlook sudden capacity loss and unpredictable demand. This study introduces a disaster-aware Traffic Assignment Problem (TAP) model that integrates a modified Bureau of Public Roads (BPR) cost function, explicitly accounting for effective capacity changes during disasters. The Frank-Wolfe (FW) algorithm is applied to solve the model, chosen for its scalability and convergence properties. A comparative analysis with Simulated Annealing (SA) is also performed across various network sizes and disruption scenarios. Results show that FW consistently delivers near-optimal flow distributions with lower travel costs and faster convergence. While SA exhibits higher variability under tight capacity constraints, FW demonstrates robust stability, particularly in medium to large networks under moderate to severe disruptions. Flow patterns from FW highlight adaptive traffic redistribution, effectively bypassing congested or blocked links. This study is the first to systematically compare Frank-Wolfe and Simulated Annealing under disaster-induced TAP conditions with capacity degradation. Contributions include (1) formulating a disaster-aware TAP model, (2) applying FW to disrupted networks, and (3) validating through structured simulations. Findings suggest that FW offers a reliable optimization tool for real-time traffic reallocation, supporting resilient urban mobility in emergencies.

Optimization of Machine Learning Models in Student Graduation Prediction Systems Using Ensemble Learning with PSO and SMOTE

Hamdani, Susanti, Lathifah, M. Khairul Anam, Rahman Pradipta — Wed, 15 Oct 2025 00:00:00 +0000

The timely graduation of students is a key metric in evaluating the academic effectiveness of higher education institutions. However, accurately identifying students at risk of delayed graduation remains challenging due to imbalanced data distributions and the instability of single-model prediction approaches. This study proposes an optimized ensemble-based machine learning system for predicting on-time graduation among university students. The model integrates C4.5, K-Nearest Neighbor (KNN), and Random Forest algorithms through a hard voting classifier, which is further optimized using Particle Swarm Optimization (PSO) to determine the most effective weighting configuration. To address class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) is implemented, ensuring balanced representation between timely and delayed graduates. A dataset of 809 student academic records from Universitas Sains dan Teknologi Indonesia (USTI) was used, and performance was evaluated using 5-fold cross-validation. The proposed ensemble model achieved an average accuracy of 93.70%, a precision of 0.94, a recall of 0.93, and an F1-score of 0.94, outperforming each individual classifier. These results confirm that the combination of ensemble learning, PSO-based optimization, and data balancing effectively improves both accuracy and model stability. The findings highlight the system’s potential as a reliable decision-support tool for educational institutions to anticipate delayed graduations and improve academic supervision strategies.

Collective Intelligence for Cybersecurity: Federated Learning under Non-IID Conditions for Intrusion Detection

Hutheifa Anwar Mohammed, Awos Kh. Ali — Thu, 02 Oct 2025 00:00:00 +0000

Cyber threats are becoming increasingly complex in cyberspace, which highlights the necessity for strong Intrusion Detection Systems (IDS). However, traditional centralized IDS methods have large problems with data privacy and scalability. Federated Learning (FL) is an intriguing new idea that lets multiple clients train a model together without sharing data directly, which keeps privacy intact. The proposed federated intrusion detection model develops and assesses FL models for detecting network intrusions, focusing on the important issue of non-independent and non-identically distributed (non-IID) data among clients. This work implements and compares two widely recognized FL algorithms, Federated Averaging (FedAvg) and Federated Proximal (FedProx), using a 1D Convolutional Neural Network (CNN) architecture specifically designed for tabular network traffic data. The authors utilize a Dirichlet distribution (α=0.1) to distribute the data among 10, 20, and 30 clients, thereby simulating non-IID conditions in the experiment. The authors thoroughly compare the performance of algorithms using two benchmark datasets: NSL-KDD and NF-Bot-Net-V2. The comparison reveals that while both FedAvg and FedProx achieve high detection rates on NSL-KDD, FedProx is more capable of maintaining stability and converging on the more complex NF-Bot-Net-V2 dataset, achieving an accuracy of 0.9953. The results highlight that FedProx is a more appropriate algorithm for implementing robust and privacy-preserving federated intrusion detection systems in statistically heterogeneous network environments found in the real world.

Implementation and Evaluation of Artificial Neural Networks for Product Sales Prediction at Basmalah Stores

Muhammad Iqbal Akkad, Mokhamad Amin Hariyadi, Agung Teguh Wibowo Almais — Wed, 22 Oct 2025 00:00:00 +0000

This study aims to develop a product sales prediction system for Toko Basmalah located in the Malang Regency area by utilizing the Artificial Neural Network (ANN) algorithm. A quantitative approach was employed, using time series sales data obtained from the Marketing Division of PT. Sidogiri Pandu Utama for the period of January 1, 2023, to December 31, 2024. The research stages included data collection and preprocessing, normalization using the min-max scaling technique, data splitting into training and testing sets, ANN model experimentation with various data compositions, and performance evaluation based on the Mean Squared Error (MSE) metric. The experiments were conducted five times using the Kaggle Editor platform. The results showed that the ANN-E model with a specific architecture achieved the lowest MSE value of 34.38%, making it the most optimal model for sales prediction. These findings are expected to assist in making better decisions regarding stock management, sales planning, and business strategies in the retail environment.

Analyzing User Acceptance of NFJuara Mobile Application Using TAM and D&M IS Success Model

Muhammad Koim, Wasilah, Chairani, Sriyanto, Sri Lestari — Thu, 02 Oct 2025 00:00:00 +0000

This study purposes to know how NFJuara application is accepted by the users in Nurul Fikri Lampung using the Technology Acceptance Model (TAM) Integrated with D&M IS Success Model. Data was collected by a validated questionnaire with inner model and outer model testing using PLS-SEM software SmartPLS. The type of data in this study is a quantitative approach. The number of samples collected was 143 respondents. Results of this research show that one of the hypotheses is rejected, that is, Service Quality (SEQ) does not affect Perceived Usefulness (PU) significantly. Besides that, this study shows that Perceived Usefulness (PU) and Perceived Ease of Use (PEU) affect as significant Acceptance of IT (AI) with R2=0.59 (Moderate) and β=0,36 (PUàAI), β=0,46 (PEUàAI). These findings imply that developers of NFJuara applications need to improve the service quality to increase acceptance, although overall NFJuara application is accepted by the user because they still feel the benefits and usefulness of the application. The contribution of this study lies in testing the technology acceptance model in the context of mobile learning, which enriches the literature on the adoption of application-based e-learning, as well as providing practical recommendations for application developers to enhance user experience.

Application of LSTM-Based Deep Learning for Stock Return Prediction of DCII

Sri Mulyani, Wanda Ilham — Wed, 22 Oct 2025 00:00:00 +0000

Stock return prediction is one of the areas that has received great attention in modern finance because it can help investors make more informed decisions and reduce the risk of market uncertainty. This study applies a deep learning approach based on Long Short-Term Memory (LSTM) to predict the return of DCII (PT DCI Indonesia Tbk) shares as a representation of highly volatile stocks on the Indonesia Stock Exchange. The purpose of this study is to evaluate the performance of twelve LSTM variants—including LSTM-Base, LSTM-Wide, LSTM-Stack2, LSTM-Stack3, LSTM with Dropout, BiLSTM, BiLSTM with Attention, and LSTM with Attention Mechanism—by comparing their performance on daily (H=1) and weekly (H=7) prediction horizons using historical data from id.investing.com. The initial data undergo preprocessing involving local format cleaning, calculation of technical indicators (MA, EMA, MACD, RSI, ATR, Bollinger Bands, etc.), MinMax normalization, and sequencing (windowing) with 30, 60, and 120-day lookbacks. The training process uses a uniform configuration with Adam optimization and early stopping to prevent overfitting, while the evaluation employs MAE, RMSE, MAPE, and R² metrics. The results show that LSTM-Stack3 (LB=60, H=1) provides the best performance with MAE = 0.020, RMSE = 0.031, MAPE = 5.0%, and R² = 0.91, followed by LSTM-Stack2-DO as the second-best configuration. Meanwhile, the LSTM-LB120-H7—the only model evaluated with a seven-day horizon—achieves the lowest performance due to higher long-term uncertainty. These findings confirm that stacked LSTM architectures are more effective for short-term return forecasting, whereas longer horizons require hybrid or enhanced approaches for stable performance..

Frequent Pattern Mining for Cyberattack Detection Using FP-Growth on Network Traffic Logs

Ali Hamsar, Fajar Maulana, Yomei Hendra, Asyahri Hadi Nasyuha, Moustafa H Aly — Thu, 02 Oct 2025 00:00:00 +0000

Cybersecurity threats have become increasingly complex, coordinated, and adaptive, creating significant challenges for traditional intrusion detection systems (IDS) that rely on static, signature-based mechanisms. These systems often fail to recognize novel, evolving, or multi-vector attacks that do not match predefined patterns. To overcome these limitations, this study proposes a data-driven framework that applies the Frequent Pattern Growth (FP-Growth) algorithm to analyze co-occurring events within network traffic logs. Using the CIC-IDS2017 benchmark dataset, which includes a wide range of real-world attack scenarios, network events were preprocessed and transformed into transactional data. This transformation enabled the efficient extraction of frequent itemsets and association rules without the computational burden of candidate generation. The experimental results show that the proposed method effectively uncovers meaningful attack correlations, such as brute force attempts preceding privilege escalation or malware infections leading to large-scale DDoS attacks. The model achieved a precision of 77.27%, recall of 70.83%, and F1-score of 73.91%, confirming its reliability in detecting sophisticated attack chains. A heatmap visualization was also generated to improve interpretability, allowing security analysts to quickly identify critical attack relationships. In conclusion, this research demonstrates that FP-Growth provides a scalable, interpretable, and computationally efficient approach to cyberattack detection, with potential integration into real-time IDS environments. Future work will focus on temporal sequence mining and hybrid models combining FP-Growth with machine learning to enhance adaptive, context-aware threat detection.

Usability and UX of Ruangguru Mobile: A 120-User Evaluation with SUS and UEQ

Cunthbert Kholin, Johanes Terang Kita Perangin Angin — Fri, 07 Nov 2025 00:00:00 +0000

The rapid growth of digital learning in Indonesia has emphasized the importance of evaluating the user interface (UI) and user experience (UX) of mobile learning platforms such as Ruangguru. This study aims to assess the usability and experiential quality of the Ruangguru mobile application to understand how effectively it supports engagement and learning motivation. A quantitative descriptive method was employed using two standardized instruments: the System Usability Scale (SUS) and the User Experience Questionnaire (UEQ). The SUS, consisting of 10 Likert-scale items, measures usability, while the UEQ evaluates six dimensions—Attractiveness, Perspicuity, Efficiency, Dependability, Stimulation, and Novelty—on a seven-point semantic differential scale. Data were collected from 120 purposively selected users (high school and university students who had used the app for at least one month and once a week) through Google Forms. Results show a mean SUS score of 76.2, classified as Good/Acceptable, exceeding the global usability benchmark of 68. UEQ results indicate Excellent ratings for Stimulation (1.89) and Novelty (1.95), and Good ratings for Attractiveness (1.61), Perspicuity (1.42), Efficiency (1.68), and Dependability (1.44). These findings highlight Ruangguru’s strong emotional engagement through gamification and visual design, though minor dependability issues suggest optimization for latency and consistency. Overall, Ruangguru demonstrates an intuitive, motivating, and functionally robust design. Future improvements should prioritize navigation flow, performance stability, and cross-device optimization. This study underscores the value of combining SUS and UEQ as a dual-instrument framework for evidence-based, user-centered design in digital learning environments.

Mobile Banking Service Quality and User Loyalty Using MSQUAL: A Systematic Literature Review

Ainun Nashikha, Muhammad Qomarul Huda, Fitroh, Yusuf Durachman, Bayu Waspodo — Thu, 02 Oct 2025 00:00:00 +0000

Digital transformation has made mobile banking a core service in the banking industry, emphasizing service quality as a critical factor for user satisfaction and loyalty. This study presents a systematic literature review (SLR) of mobile banking research from 2021 to 2025, guided by PRISMA and structured using the PICOC framework (Population, Intervention, Comparison, Outcome, Context) to systematically select and evaluate relevant studies. The MS-QUAL model, comprising nine dimensions: efficiency, system availability, responsiveness, privacy, content, contact, billing, fulfillment, and compensation, was used as the evaluation framework. Out of 924 initially identified articles, 20 met the inclusion criteria for in-depth analysis. Findings show that efficiency, system availability, privacy, responsiveness, content, and fulfillment consistently drive user satisfaction, while compensation, contact, and billing have limited influence. Satisfaction serves as the primary mediator connecting service quality to loyalty, indicating that improvements in MS-QUAL dimensions must translate into positive user experiences to foster long-term loyalty. The study further highlights challenges in maintaining security standards, adapting traditional dimensions to evolving user expectations, and ensuring consistent service quality. Opportunities lie in leveraging technologies such as AI, blockchain, and big data to create personalized, secure, and interactive experiences, enhancing both functional and emotional engagement. Overall, MS-QUAL remains a relevant and flexible framework for evaluating mobile banking service quality when aligned with contemporary technological advances and user-centered strategies.

Integration of Machine Learning and Blockchain for Forest Fire Risk Prediction

Nursetiaji Ramadhani, Ira Diana Sholihati, Djarot Hindarto, Asrul Sani — Mon, 10 Nov 2025 00:00:00 +0000

This study presents an integrated framework combining machine learning and blockchain technology to enhance the accuracy, transparency, and reliability of forest fire risk prediction in tropical regions. Using geospatial and climatological datasets from Google Earth Engine (GEE), two ensemble algorithms—Random Forest (RF) and Extreme Gradient Boosting (XGBoost)—were trained to model spatial fire susceptibility based on variables such as temperature, humidity, rainfall, wind speed, and vegetation index (NDVI). The RF model effectively identified low-risk areas but was less sensitive to minority high-risk classes, while XGBoost demonstrated superior adaptability in handling class imbalance and achieved more balanced performance across all categories. To ensure data authenticity and traceability, the prediction results were validated and recorded on the Ethereum blockchain using smart contracts. Each prediction output was transformed into a cryptographic hash (SHA-256) to guarantee immutability and verifiability. The integration of machine learning with blockchain establishes a decentralized, tamper-proof, and verifiable prediction system, promoting data integrity and transparency in environmental monitoring. Overall, this research introduces a novel “verifiable prediction pipeline” that advances both artificial intelligence and blockchain applications in environmental informatics, supporting proactive and accountable forest fire mitigation strategies.

Smart CRM Application Development Using Artificial Intelligence and Extreme Programming Method

Doni Syofiawan, Miftahul Ilmi — Thu, 02 Oct 2025 00:00:00 +0000

Customer Relationship Management (CRM) is an important strategy for companies to understand customer behavior, increase loyalty, and reduce churn rates. However, the challenge that is often faced is how to manage increasingly complex customer transaction data and turn it into useful information for decision-making. This research aims to develop an artificial intelligence-based smart CRM application by integrating the K-Means algorithm for customer segmentation and XGBoost for retention prediction, as well as using the Extreme Programming (XP) methodology in the development process. The XP methodology was chosen because it is able to provide a fast, adaptive, and user-oriented iterative cycle, so that applications can be developed according to user needs. The results showed that K-Means can group customers into segments that are relevant to marketing strategies, while XGBoost provides retention prediction results with good accuracy. In addition, the application was tested using Blackbox Testing to ensure that the functionality runs according to specifications, as well as the System Usability Scale (SUS) which resulted in an average score of 89 and was included in the excellent usability category. This confirms that the system built is not only technically feasible, but also well received by users. This research contributes to presenting a smart CRM application that combines AI with modern software development methodologies, as well as opening up opportunities for advanced research at a larger data scale and integration with digital marketing systems.

Hybrid Artificial Intelligence–Blockchain Approach for Landslide Risk Classification and Recommendation

Rizal Indriawan, Ratih Titi Komalasari, Djarot Hindarto, Asrul Sani — Sun, 02 Nov 2025 00:00:00 +0000

Increased rainfall intensity, steep topography, and changes in land use in Indonesia, particularly in Java, such as Garut Regency, have increased the risk of landslides that have a widespread impact on public safety and environmental stability. This study proposes a Hybrid Artificial Intelligence and Blockchain approach to develop an accurate, secure, and transparent landslide risk classification and recommendation system. The model integrates three Multi-Criteria Decision Making (MCDM) methods, namely Analytic Hierarchy Process (AHP), Technique for Order Preference by Similarity to Ideal Solution (TOPSIS), and VlseKriterijumska Optimizacija I Kompromisno Resenje (VIKOR). These three methods are used sequentially to determine criterion weights, calculate ideal solutions, and produce optimal compromise decisions based on geospatial factors. The dataset used consists of 766 geospatial observation data covering stability, rainfall, vegetation, river distance, slope, prediction, and ground truth parameters, obtained from satellite data and open geospatial repositories in the Java Island region. The research process included pre-processing, normalization, weighting analysis using AHP–TOPSIS–VIKOR, and integration of the results into the Ethereum Blockchain Smart Contract system with a Proof of Authority (PoA) consensus mechanism. The test results showed a 17.8% increase in classification accuracy and a 21.4% increase in data storage efficiency compared to conventional methods. This approach is expected to improve the reliability, security, and transparency of the analysis system and mitigate the risk of landslides based on smart technology in Indonesia.

Comparative Performance Benchmarking of WebSocket Libraries on Node.js and Golang

Louis Fernando, Mychael Maoeretz Engel — Thu, 02 Oct 2025 00:00:00 +0000

The demand for responsive real-time web applications continues to grow, making the selection of backend technology and WebSocket libraries a crucial factor in determining performance. Node.js and Golang are popular platforms for real-time applications. However, the WebSocket library within them offers a trade-off between features and efficiency, the impact of which has not been comprehensively measured. This research aims to fill this gap by conducting a quantitative performance analysis to compare the efficiency and scalability of four WebSocket libraries: ws and socket.io on Node.js, and gorilla/websocket and coder/websocket on Golang. This research uses a benchmarking experimental method with client load simulations that gradually increase from 100 to 1000 concurrent clients. The experiment was conducted through two scenarios, namely the Echo Test and Broadcast Test. In the Echo Test, the performance metrics measured were Connection Time, Round Trip Time (RTT), and Throughput. Meanwhile, in the Broadcast Test, the performance metric measured was Broadcast Latency. The results from the Echo Test show a significant performance disparity. At a peak load of 1000 clients, socket.io achieved a throughput of only 27,152 messages/second, whereas the lightweight libraries (ws, gorilla/websocket, and coder/websocket) all achieved over 44,000 messages/second. In the Broadcast Test with a high load, the latency difference between the four libraries became insignificant. Therefore, for applications prioritizing raw performance in point-to-point communication, certain WebSocket libraries such as ws, gorilla/websocket, and coder/websocket are more suitable for future development.

Naïve Bayes–Based Chatbot with Sentiment Analysis for Culinary Preferences in Bali

Anak Agung Sandatya Widhiyanti, I Gusti Agung Ayu Sekarini — Thu, 02 Oct 2025 00:00:00 +0000

The rapid growth of digital technology has increased the adoption of chatbots across industries, including the culinary and tourism sectors. However, existing systems often lack integration of customer sentiment and user preferences, limiting recommendation relevance. This study develops a personalized chatbot by combining sentiment analysis of Google Maps reviews with user taste preferences for traditional Balinese cuisine. A dataset of 5,000 reviews was analyzed using the Naïve Bayes classifier, achieving 88% accuracy. User evaluation with 100 respondents showed positive perceptions of usability and engagement, though recommendation suitability scored lower. The findings highlight the potential of sentiment-driven personalization and suggest future improvements through advanced models, larger datasets, and multilingual features for tourism.

Designing a Stunting Prediction Model Using Machine Learning to Support SDGs Achievement in Indonesia

Mikha Sinaga, Fujiati, Darma Halawa — Fri, 03 Oct 2025 00:00:00 +0000

Stunting remains a major public health challenge in Indonesia, with national prevalence among children under five reaching 21.6% in 2022, according to the Ministry of Health. This condition, defined by the World Health Organization as a height-for-age less than -2 SD, is associated with long-term consequences including impaired cognitive development, reduced educational attainment, and diminished economic productivity. Addressing stunting is therefore critical to achieving Sustainable Development Goals (SDGs) related to hunger, health, and education. Despite multiple national initiatives, early identification of stunting risk is still limited by reliance on conventional, reactive surveillance methods. Recent advances in machine learning (ML) provide promising alternatives for proactive stunting prediction, with several studies reporting high predictive accuracy using ensemble methods, hybrid frameworks, and geographically weighted models. Building upon this evidence, the present study develops and evaluates ML models for stunting risk prediction using a large dataset of 10,000 records from North Sumatra, Indonesia. The dataset included three predictor variables—age, height, and weight—and a target variable, nutritional status (Normal, Stunted, Severely Stunted, Tall). Four algorithms were compared: K-Nearest Neighbors (KNN), Naïve Bayes, Decision Tree, and Random Forest. Performance was assessed using accuracy, precision, recall, F1-score, and ROC area, with 10-fold cross-validation ensuring robust estimation. Results demonstrated that Decision Tree (88.6% accuracy) and Random Forest (88.3% accuracy) outperformed KNN (84.7%) and Naïve Bayes (72%). ROC areas further confirmed the superiority of ensemble-based approaches, particularly Random Forest (0.979). Statistical significance was tested using McNemar’s test, revealing that Decision Tree and Random Forest achieved comparable performance (p = 0.651), both significantly outperforming KNN and Naïve Bayes (p < 0.05). This study contributes a context-specific evaluation of ML methods for stunting prediction in North Sumatra, emphasizing not only predictive accuracy but also interpretability to support health policy and program implementation. By bridging data-driven insights with actionable decision support, the proposed framework advances progress toward SDG-aligned strategies and provides a foundation for more targeted and preventive interventions in child nutrition and growth monitoring.

A Hybrid PULTS–SWARA–ELECTRE-I Model for Multi-Criteria Political Sentiment Classification on Indonesian Twitter Data

Taufik Yuandika Bahri, Chairini Aisyah — Wed, 15 Oct 2025 00:00:00 +0000

Social media platforms such as Twitter have become crucial for analyzing political sentiment, particularly in contexts where public opinion shifts rapidly. This study proposes a hybrid classification model that combines Probabilistic Uncertain Linguistic Term Set (PULTS), Stepwise Weight Assessment Ratio Analysis (SWARA), and ELimination Et Choice Translating REality (ELECTRE-I). Using a dataset of 7,800 tweets collected between January and July 2024 covering five major political parties in Indonesia, the model classifies tweets into positive, negative, and neutral sentiments. To address class imbalance, Easy Data Augmentation (EDA) was applied, while Term Frequency–Inverse Document Frequency (TF-IDF) was used for feature extraction. The results show that the proposed model achieves 90% accuracy and an F1-score of 85%, outperforming baseline methods such as SVM (86.7%), Naïve Bayes (83.3%), Decision Tree (88%), and K-Means (76.7%). These improvements demonstrate that the integration of linguistic uncertainty with expert-driven feature weighting provides measurable advantages in political sentiment classification. Beyond performance, the study contributes theoretically by extending multi-criteria decision-making methods into sentiment analysis and by offering a more interpretable alternative to opaque machine learning models. Together, these findings highlight the practical value of explainable decision frameworks for political communication while advancing methodological approaches for analyzing sentiment under uncertainty.

Cervical Cancer Classification Using Multi-Directional GLCM Shape-Texture Features in LBC

Surmayanti, Irohito Nozomi, Febri Aldi — Fri, 10 Oct 2025 00:00:00 +0000

Alsalatie, M., Alquran, H., Mustafa, W. A., Zyout, A., Alqudah, A. M., Kaifi, R., & Qudsieh, S. (2023). A New Weighted Deep Learning Feature Using Particle Swarm and Ant Lion Optimization for Cervical Cancer Diagnosis on Pap Smear Images. Diagnostics, 13(17), 2762. https://doi.org/10.3390/diagnostics13172762

Arbyn, M., Weiderpass, E., Bruni, L., Sanjosé, S. de, Saraiya, M., Ferlay, J., & Bray, F. (2020). Estimates of incidence and mortality of cervical cancer in 2018: A worldwide analysis. The Lancet Global Health, 8(2), e191–e203. https://doi.org/10.1016/S2214-109X(19)30482-6

Attallah, O. (2023). Cervical Cancer Diagnosis Based on Multi-Domain Features Using Deep Learning Enhanced by Handcrafted Descriptors. Applied Sciences, 13(3), 1916. https://doi.org/10.3390/app13031916

Chaddad, A., & Tanougast, C. (2017). Texture Analysis of Abnormal Cell Images for Predicting the Continuum of Colorectal Cancer. Analytical Cellular Pathology, 2017(1), 8428102. https://doi.org/10.1155/2017/8428102

Díaz del Arco, C., & Saiz Robles, A. (2024). Advancements in Cytological Techniques in Cancer. In Handbook of Cancer and Immunology (pp. 1–46). Springer, Cham. https://doi.org/10.1007/978-3-030-80962-1_385-1

Garg, M., & Dhiman, G. (2021). A novel content-based image retrieval approach for classification using GLCM features and texture fused LBP variants. Neural Computing and Applications, 33(4), 1311–1328. https://doi.org/10.1007/s00521-020-05017-z

Huang, X., Liu, X., & Zhang, L. (2014). A Multichannel Gray Level Co-Occurrence Matrix for Multi/Hyperspectral Image Texture Representation. Remote Sensing, 6(9), 8424–8445. https://doi.org/10.3390/rs6098424

Ikeda, K., Oboshi, W., Hashimoto, Y., Komene, T., Yamaguchi, Y., Sato, S., Maruyama, S., Furukawa, N., Sakabe, N., & Nagata, K. (2021). Characterizing the Effect of Processing Technique and Solution Type on Cytomorphology Using Liquid-Based Cytology. https://dx.doi.org/10.1159/000519335

Kaur, H., Sharma, R., & Kaur, J. (2025). Comparison of deep transfer learning models for classification of cervical cancer from pap smear images. Scientific Reports, 15(1), 3945. https://doi.org/10.1038/s41598-024-74531-0

Merlina, N., Noersasongko, E., Nurtantio, P., Soeleman, M. A., Riana, D., & Hadianti, S. (2021). Detecting the Width of Pap Smear Cytoplasm Image Based on GLCM Feature. In Y.-D. Zhang, T. Senjyu, C. SO–IN, & A. Joshi (Eds.), Smart Trends in Computing and Communications: Proceedings of SmartCom 2020 (pp. 231–239). Springer. https://doi.org/10.1007/978-981-15-5224-3_22

Mishra, G. A., Pimple, S. A., & Shastri, S. S. (2021). An overview of prevention and early detection of cervical cancers. Indian Journal of Medical and Paediatric Oncology, 32, 125–132.

Plissiti, M. E., Nikou, C., & Charchanti, A. (2011). Combining shape, texture and intensity features for cell nuclei extraction in Pap smear images. Pattern Recognition Letters, 32(6), 838–853. https://doi.org/10.1016/j.patrec.2011.01.008

Raga Permana, Z. Z., & Setiawan, A. W. (2024). Classification of Cervical Intraepithelial Neoplasia Based on Combination of GLCM and L*a*b* on Colposcopy Image Using Machine Learning. 2024 International Conference on Artificial Intelligence in Information and Communication (ICAIIC), 035–040. https://doi.org/10.1109/ICAIIC60209.2024.10463256

Rastogi, P., Khanna, K., & Singh, V. (2023, August 8). Classification of single‐cell cervical pap smear images using EfficientNet—Rastogi—2023—Expert Systems—Wiley Online Library. https://onlinelibrary.wiley.com/doi/abs/10.1111/exsy.13418

Singh, D., Vignat, J., Lorenzoni, V., Eslahi, M., Ginsburg, O., Lauby-Secretan, B., Arbyn, M., Basu, P., Bray, F., & Vaccarella, S. (2023). Global estimates of incidence and mortality of cervical cancer in 2020: A baseline analysis of the WHO Global Cervical Cancer Elimination Initiative. The Lancet Global Health, 11(2), e197–e206. https://doi.org/10.1016/S2214-109X(22)00501-0

Singh, T. G., & Karthik, B. (2023). Accurate Cervical Tumor Cell Segmentation and Classification from Overlapping Clumps in Pap Smear Images. In S. N. Singh, S. Mahanta, & Y. J. Singh (Eds.), Proceedings of the NIELIT’s International Conference on Communication, Electronics and Digital Technology (pp. 659–673). Springer Nature. https://doi.org/10.1007/978-981-99-1699-3_46

Strander, B., Andersson-Ellström, A., Milsom, I., Rådberg, T., & Ryd, W. (2007). Liquid-based cytology versus conventional Papanicolaou smear in an organized screening program. Cancer Cytopathology, 111(5), 285–291. https://doi.org/10.1002/cncr.22953

Wahidin, M., Febrianti, R., Susanty, F., & Hasanah, S. R. (2022, March 1). Twelve Years Implementation of Cervical and Breast Cancer Screening Program in Indonesia—PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC9360967/

ECG-Based Heart Rate Variability and KNN Classification for Early Detection of Baby Blues Syndrome in Postpartum Mothers

Thu, 02 Oct 2025 00:00:00 +0000

Early detection of baby blues syndrome plays an important role in preventing postpartum emotional disturbances from developing into more serious mental health conditions. This study proposes a simple and non-invasive approach to identify early signs of baby blues in postpartum mothers by analyzing electrocardiogram (ECG) signals using the K-Nearest Neighbor (KNN) algorithm. The ECG data were gathered through wearable sensors and processed to extract heart rate variability (HRV) features such as RMSSD, SDNN, entropy, and energy. These features were then used to train and test a KNN classification model through a five-fold cross-validation process. KNN was chosen because it is easy to implement, does not assume any specific data pattern, and works well with small datasets like those commonly found in clinical settings. Its ability to group data based on similarity makes it suitable for recognizing subtle physiological changes linked to emotional stress. The model reached an accuracy of 87.5%, with strong precision and recall scores, showing its reliability in distinguishing mothers who show early symptoms of baby blues from those who do not. Among all features, RMSSD and SDNN had the highest impact, pointing to reduced parasympathetic activity in affected individuals. These findings suggest that combining HRV analysis with a straightforward machine learning approach like KNN offers a promising, low-cost solution for early emotional screening in maternal care, especially where resources are limited.

Comparative Analysis of SDLC and R&D Methods in System Development: A Case Study of Integrity Zone Management System

Adidtya Perdana, Sri Dewi, Nurul Ain Farhana, Didi Febrian — Sat, 18 Oct 2025 00:00:00 +0000

This paper presents a comprehensive comparative analysis of Software Development Life Cycle (SDLC) and Research and Development (R&D) methodologies in system development, with a specific focus on their application to the Integrity Zone Management Information System. Through a systematic literature review and an in-depth case study analysis, this research examines the fundamental differences, strengths, and limitations of each methodology. The study identifies key dimensions for comparison including flexibility, risk management, innovation potential, documentation requirements, and stakeholder engagement. Findings reveal that while SDLC methodologies provide structure and predictability for well-defined requirements, R&D approaches offer greater innovation capacity for exploratory projects. The Integrity Zone Management Information System case demonstrates how hybrid approaches can leverage the strengths of both methodologies and improved stakeholder satisfaction by 94%. This research contributes to the theoretical understanding of system development methodologies and provides practical guidance for selecting appropriate approaches based on project context, objectives, and constraints. The paper concludes with recommendations for practitioners and suggestions for future research in methodological integration and adaptation.

Real-Time Web-Based Ship Collision Risk Detection Using AIS Data and Collision Risk Index (CRI)

Fri, 03 Oct 2025 00:00:00 +0000

The high density of maritime traffic in Indonesian waters, particularly in the Lombok Strait and Nusa Penida region, increases the risk of ship collisions, especially among vessels lacking adequate navigation systems. This study presents the development of a web-based system for real-time ship monitoring and collision risk assessment using Automatic Identification System (AIS) data. The system integrates a backend powered by FastAPI and MongoDB with a frontend built using React JS. AIS data is collected from a base station and processed to detect ship encounters using the DBSCAN clustering algorithm combined with Haversine distance to identify encounter detection. The risk assessment applies the Collision Risk Index (CRI) method by calculating DCPA (Distance to Closest Point of Approach) and TCPA (Time to Closest Point of Approach), allowing for graded risk categorization. Real-time risk notifications are delivered via WebSocket, and the interface includes interactive maps, ship detail views, and maritime weather information from the BMKG API. The system achieved high responsiveness, with an average detection time of 0.0075 seconds per ship and an end-to-end response time of approximately 61 milliseconds. Functional and usability tests show that the system effectively supports early detection of collision risks and improves maritime situational awareness. The proposed solution is scalable and applicable for maritime safety monitoring in busy sea routes, contributing to safer navigation and proactive decision-making.

Integration Of Pca And K-Means Clustering For Staple Food Segmentation In Support Of National Food Policy

Sardo Sipayung, Paska Marto Hasugian — Fri, 17 Oct 2025 00:00:00 +0000

This study aims to develop cross-provincial staple-food segmentation by integrating Principal Component Analysis (PCA) and K-Means to support policy formation. The dataset comprises 2023 staple-food consumption for 34 Indonesian provinces across six indicators from BPS/SUSENAS. All indicators were standardized using z-score, reduced via PCA, and the resulting component scores were used as inputs to K-Means. Three components (PC1–PC3) explained 73.86% of the variance and captured shifts between sweet/animal-based vs. plant foods, fatty or animal-based grains, and the energy contribution of fat. The optimal number of clusters was determined as k = 3, yielding Silhouette = 0.466 and DBI = 0.733, indicating sufficiently compact and well-separated groups. The results reveal three segments: the first group consists of 11 provinces that are predominantly plant-based with low sugar and low animal-based consumption; the second group includes 13 provinces characterized by high animal-based and high-fat consumption; and the third group comprises 10 provinces with low-fat diets and fresh plant-based consumption. Stability checks on initialization and a leave-one-feature-out procedure confirmed consistent assignments. This fills an empirical gap: to our knowledge, no prior research integrates PCA with K-Means for cross-provincial staple-food segmentation in Indonesia while also reporting internal validation. Practically, the study provides operational segmentation to support food-security interventions moving beyond composite indices toward actionable targeting for production support, supply/price stabilization, and improved nutritional access thereby reframing IKP/FSVA from index-ranking to evidence-based segmentation.

Enterprise Architecture for the Cruise Industry: A TOGAF-ADM and ArchiMate-Based Approach

William Watasendjaja, Glenny Chudra, Alfa Yohannis — Thu, 02 Oct 2025 00:00:00 +0000

Despite its growth and resilience over the last decades, the cruise industry faces significant challenges in its strategic, operational, and technology domains. The unique complexity of the industry requires cruise companies to adopt a structured approach to enterprise transformation. To address this problem, this aims to provide an Enterprise Architecture (EA) blueprint for the cruise industry. Using a case study of a leading cruise line, CruiseX, this study analyzes the operational model of the cruise line and apply two industry-leading standards: The Open Group Architecture Framework (TOGAF) and the ArchiMate modelling language. This study applies the four core phases of TOGAF Architecture Development Method (ADM) from the initial phase of Architecture Vision (Phase A), through the definition of Business Architecture, Information System Architecture, and Technology Architecture (Phase B to D). The ArchiMate language is utilized to visualize the core business processes, information systems, and technology architecture. By using TOGAF ADM as the technical guidelines and ArchiMate as the modeling language, the result of this study is a blueprint of core business processes, application and data that support each business processes, and the underlying technology infrastructure, that provides a structured framework and serves as an actionable tool for implementing enterprise architecture in cruise industry. This research also extends the application of TOGAF and ArchiMate to the under-research cruise industry domain. The study’s limitations include the reliance on publicly available data, the limited scope of business processes, and the lacks of practitioner validation, suggesting clear directions for future research.

Hybrid CNN and KNN Approach for Coffee Bean Quality Identification

Widya Lelisa Army, Sri Anita, Retno Ramadhina — Sun, 26 Oct 2025 00:00:00 +0000

This study discusses the integration of Convolutional Neural Network (CNN) and K-Nearest Neighbors (KNN) for the identification of coffee bean quality as an effort to increase the competitiveness of local commodities. CNN is used as a feature extractor to produce an information-rich representation of coffee bean images, while KNN acts as a classifier to classify quality into two classes, namely Good and Defective. The dataset is divided into training, validation, and test data, with a total of 1,190 images obtained from the manual annotation process. The research stages include (1) pre-processing of data in the form of cropping based on bounding boxes, resize to 224×224 pixels, normalization, and data augmentation; (2) feature extraction using pretrained CNN (ResNet18) by eliminating the final classification layer to obtain a 512-dimensional feature vector; and (3) classification using KNN with variations in k values (3, 5, and 7) as well as Euclidean distance metrics. The results of the experiment showed that the CNN+Softmax baseline resulted in an accuracy of 86%, while the CNN+KNN method provided better performance. The k=5 configuration was proven to be optimal with an accuracy of 93.4%, precision, recall, and a balanced F1-score in both classes. The confusion matrix shows that most samples can be classified correctly with a low error rate. These findings are in line with previous research that emphasized the effectiveness of CNN in the extraction of visual features and the advantages of KNN on limited datasets. Thus, this approach can be a practical solution to support an automatic, accurate, and consistent coffee bean quality identification system to increase the competitiveness of local coffee commodities in the global market.

A Comparative Study between Logistic Regression and SVM for Resource Management in Network Slicing

Ahmed Younus, Ali Al-Allawee — Thu, 02 Oct 2025 00:00:00 +0000

Network slicing is an essential component of 5G and subsequent networks. It enables administrators to partition shared physical infrastructure into several virtual segments, each with distinct Quality of Service (QoS) requirements. Effective and adaptable real-time resource management is essential for optimal performance in dynamic situations, characterized by low latency and high throughput. Despite the increasing body of literature on machine learning in communication networks, there is a paucity of direct comparisons between Logistic Regression (LR) and Support Vector Machines (SVM) concerning network slicing resource management. Prior comparisons have predominantly concentrated on sectors such as education, healthcare, and the Internet of Things (IoT), resulting in minimal exploration of slicing prospects. This study rectifies this gap by doing a comparative analysis of Logistic Regression and Support Vector Machine models utilizing the CICIDS2017 dataset in a network slicing simulation environment. Both models were utilized independently, employing class balancing and feature selection to forecast overload. We evaluated their performance for accuracy, ROC AUC, latency, jitter, and throughput across network slices. Results indicate that SVM exhibited somewhat superior classification accuracy; however, LR consistently surpassed SVM in critical network-level parameters, including reduced delay, enhanced throughput, and improved jitter stability. These results indicate that LR is an effective option for the real-time management of network slicing resources due to its practicality and comprehensibility. In conclusion, LR is a dependable primary option for scholars and professionals pursuing effective, low-latency solutions, improving the superior classification accuracy of SVM with enhanced overall network performance.

Comparative Performance Evaluation of MobileNetV3 and ResNet50 for Forest Fire Image Classification

Muhammad Rizky Amirullah Hidayat, Djarot Hindarto, Asrul Sani — Wed, 05 Nov 2025 00:00:00 +0000

Indonesia is one of the countries with a high incidence of forest and land fires (karhutla), especially during the dry season, thus requiring a fast and efficient early detection system. This study aims to compare the performance of two popular deep learning architectures, namely MobileNetV3 (Large and Small variants) and ResNet50, in forest fire image classification tasks using a transfer learning-based approach. This study emphasizes the comparison between accuracy and computational efficiency in a CPU-only environment, which represents real-world conditions of use in the field without GPU support. The dataset used is a combination of local field images from the Puncak area, Bogor, and a curated public forest fire dataset to ensure the model's generalization ability to diverse geographical conditions. The results of the experiment show that ResNet50 provides the highest accuracy with a training accuracy value of 0.677 and a validation accuracy of 0.647, but requires longer training and inference times. Meanwhile, MobileNetV3-Large and MobileNetV3-Small showed better computational efficiency, with only slightly lower accuracy (0.635 and 0.61) and high training stability. These findings confirm that lightweight models such as MobileNetV3 strike an optimal balance between accuracy, speed, and resource consumption, making them an ideal solution for implementing edge computing-based early detection systems. Overall, this research contributes by providing an empirical comparative analysis that can serve as a reference for selecting deep learning architectures for efficient and adaptive forest fire detection systems that are constrained by hardware limitations.

Blockchain Model for Tracking Plastic Waste Using Smart Contracts to Reduce Emissions

Andri Dwi Utomo, Jeffry, Ahmad Irfandi — Thu, 02 Oct 2025 00:00:00 +0000

This research focuses on the design and development of a blockchain-based plastic waste tracking system aimed at enhancing transparency, efficiency, and accountability in plastic waste management. The system utilizes Hyperledger Fabric as a permissioned blockchain platform and integrates smart contracts to manage transactions between organizations, including waste generators, collectors, sorting warehouses, and final processing warehouses. This system records each stage of the plastic waste journey, from creation to final processing, in a permanent, transparent, and immutable manner. The testing results demonstrate that the system can accurately record the status and history of waste, manage transfers between organizations, and process plastic waste into recycled products. Moreover, the system shows a significant potential for carbon emission reduction, with an estimated reduction of up to 50% compared to traditional plastic waste management methods, such as incineration or landfilling. The study also explores how the implementation of blockchain can support global efforts in mitigating the environmental impacts of plastic waste. The blockchain-based system also provides real-time monitoring, ensuring that each transaction is verified and recorded immediately, contributing to more effective management. The implementation of smart contracts further guarantees that waste-related activities are executed automatically when predefined conditions are met, reducing administrative overhead. The study also explores how the implementation of blockchain can support global efforts in mitigating the environmental impacts of plastic waste. Ultimately, this system presents a scalable solution that could be adopted in various regions to improve global waste management strategies.

MCDM-based Fire Risk Mapping with Geospatial Visualization and Blockchain

Emmanuel Abet Rossi Paays, Djarot Hindarto, Asrul Sani — Sun, 02 Nov 2025 00:00:00 +0000

Forest fires are among the most destructive environmental disasters in Indonesia, causing long-term ecological damage, health problems, and economic disruption. Increasing occurrences driven by climate anomalies, land clearing, and vegetation dryness highlight the need for intelligent and data-driven risk monitoring systems. This study introduces a hybrid analytical framework that integrates Multi-Criteria Decision-Making (MCDM) with blockchain-based data management and geospatial visualization to identify forest fire risk levels. The proposed model combines the Analytic Hierarchy Process (AHP), Weighted Sum Model (WSM), and Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) to evaluate multiple parameters, including temperature, humidity, rainfall, and the Normalized Difference Vegetation Index (NDVI). Environmental data were securely obtained from a private Ethereum blockchain using Ganache, Truffle, and MetaMask to ensure transparency, integrity, and immutability. Results were visualized through an interactive Leaflet.js interface, allowing real-time geospatial monitoring linked to blockchain transaction hashes. The AHP analysis revealed that temperature (0.36) and humidity (0.27) contributed 63% of the total decision weight, while TOPSIS identified high-risk zones consistent with historical records. Validation against BNPB data achieved 90.7% accuracy, confirming the model’s reliability. The integration of MCDM, GIS, and blockchain provides a transparent, decentralized, and verifiable approach for national-scale fire-risk management, enhancing the accuracy and credibility of environmental decision-making systems.

Fairer Public Complaint Classification on LaporGub: Integrating XLM-RoBERTa with Focal Loss for Imbalance Data

Thu, 02 Oct 2025 00:00:00 +0000

The advancement of digital technology has provided opportunities for governments to improve the quality of public services through citizen complaint channels. One example of this implementation in Indonesia is Lapor Gub, managed by the Dinas Komunikasi dan Informasi Provinsi Jawa Tengah (Communication and Information Agency of Central Java Province). This platform receives thousands of complaints daily, ranging from infrastructure, social issues, to illegal levies. However, the large volume of data and the imbalanced distribution of categories pose significant challenges for both manual and automated processing. This study aims to classify citizen complaint texts using XLM-RoBERTa combined with Focal Loss as an approach to handle data imbalance. The dataset consists of 53,774 complaints after data cleaning and text preprocessing. The training process applied a stratified split (78% training, 18% validation, 10% testing) and fine-tuning for 10 epochs. Model performance was evaluated using accuracy, precision, recall, and macro F1-score. The results show that the model without Focal Loss achieved 78.1% accuracy with a macro F1-score of 0.606, while the model with Focal Loss improved the macro F1-score to 0.625 with 78.5% accuracy. These findings demonstrate that the application of Focal Loss enhances the model’s ability to recognize minority categories without reducing performance on majority classes. Therefore, the combination of RoBERTa and Focal Loss offers an effective solution to support faster, fairer, and more transparent public complaint management.

Sentiment Analysis of Roblox Game Reviews Using Support Vector Machine Method

Thu, 02 Oct 2025 00:00:00 +0000

The development of digital technology has driven changes in entertainment consumption patterns, especially among the younger generation. Roblox has become one of the most popular online gaming platforms, with a wide range of user opinions recorded on Google Play Store. This study aims to classify the sentiment of Roblox user reviews (positive, negative, neutral) and evaluate the performance of the Support Vector Machine (SVM) algorithm with TF-IDF weighting and automatic labeling using Lexicon InSet. Data was obtained by crawling 10,000 reviews during the period of April 2–May 23, 2025, and after the preprocessing stage, 8,950 data remained for analysis. The classification results show that the sentiment distribution consists of 41.3% positive (3,703 reviews), 41.8% neutral (3,739 reviews), and 16.8% negative (1,507 reviews). Model evaluation using a confusion matrix produced high performance with 87.03% accuracy, 87.29% precision, 87.03% recall, and an F1-score of 86.67%. WordCloud visualization shows that positive reviews emphasize creativity and interactive features, while negative reviews are dominated by technical complaints such as lag and errors. These findings prove that the combination of SVM, TF-IDF, and Lexicon InSet is effective in sentiment analysis and provides valuable input for developers to improve application quality and user protection. Further research is recommended to adopt a hybrid approach based on deep learning and aspect-based sentiment analysis to generate more insights.