Predicting smartphone addiction risk based on behavioral indicators using machine learning

Objectives The rapid proliferation of mobile technologies has exponentially increased smartphone reliance among youth, turning excessive use into a critical contemporary concern that negatively impacts well-being. This study aimed to predict smartphone addiction risk in university students using behavioral, demographic, and usage indicators through interpretable machine learning (ML) methods. Methods Data from 304 valid student responses were analyzed using the Smartphone Addiction Scale-Short Version (SAS-SV) and a demographic questionnaire. A binary risk label was established based on eight behavioral indicators (mean: 2.85). A cut-off threshold of 3.0 classified students into LowRisk (less than 3.0) or AtRisk (greater than or equal to 3.0). Six ML algorithms—Decision Tree, Efficient Logistic Regression (ELR), Kernel Naive Bayes, K-Nearest Neighbours, Linear Discriminant Analysis, and Linear Support Vector Machine (LSVM)—were evaluated using 10-fold cross-validation and Chi-square feature selection. SHAP analysis enhanced the best-performing model’s interpretability. Results The LSVM and ELR models demonstrated the most balanced performance, achieving an accuracy of approximately 77%. SHAP analysis revealed that feedback from the individual’s social circle regarding their smartphone usage, along with physical discomfort associated with phone use, were the most significant variables explaining addiction risk. Conclusion These results suggest that interpretable machine learning methods can successfully identify individuals at risk using demographic and behavioral metrics, serving as effective decision-support tools for early intervention in smartphone addiction.