Predictive Modeling of Smartphone Addiction: Performance Evaluation of KNN, XGBoost, and Naive Bayes on Behavioral Patterns

Abstract

Excessive smartphone use has triggered a global crisis in the form of smartphone addiction, which negatively impacts mental health and productivity. Most current detection methods still rely on subjective questionnaires that are prone to bias. Therefore, this study aims to evaluate and compare the performance of machine learning-based predictive models—namely K-Nearest Neighbors (KNN), Naive Bayes, and Extreme Gradient Boosting (XGBoost)—in objectively classifying addiction levels based on user behavioral patterns. The research methodology adopts a standard machine learning workflow encompassing data preprocessing, model training, and performance evaluation using a dataset of 3,300 user activity log entries. Empirical results demonstrate that XGBoost yields superior classification performance, achieving an accuracy of 97.27% and an F1-Score of 96.70%, significantly outperforming the KNN (94.54%) and Naive Bayes (89.09%) algorithms. Further feature importance analysis confirms that App Usage Time is the most crucial predictor in detecting addiction. This study concludes that the XGBoost architecture is highly robust in handling non-linear behavioral feature interactions, enabling high-precision predictions. The implications of these findings provide a solid technical foundation for the development of automated early detection systems. Future research should consider expanding the dataset to include application categorization and integrating XGBoost modeling into real-time digital wellbeing application prototypes.

GS Cited Analysis