<![CDATA[The increasing use of social media and digital platforms has raised concerns regarding its potential relationship with sleep patterns, lifestyle behaviors, productivity, and psychological well-being. Stress is a common health-related issue that may be influenced by daily behavioral patterns, including screen time, social media usage, sleep duration, physical activity, and work or study habits. This study aims to develop and evaluate machine learning models for predicting stress levels based on non-invasive digital behavior and lifestyle indicators. The dataset used in this study consisted of 11,000 records with three stress level categories: Low, Medium, and High. The predictor variables included age, daily screen time, social media usage duration, sleep hours, exercise duration, study or work hours, productivity score, and the most frequently used social media platform. Several machine learning algorithms were evaluated, including Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, K-Nearest Neighbors, and Gradient Boosting. Model performance was assessed using accuracy, precision, recall, F1-score, confusion matrix analysis, and 5-fold stratified cross-validation. The experimental results showed that the overall classification performance was modest. The Decision Tree model achieved the best testing performance with an accuracy and macro F1-score of 0.3400, while Gradient Boosting achieved the highest cross-validation performance with a mean accuracy of 0.3480 and a mean macro F1-score of 0.3467. Feature importance analysis using Random Forest indicated that productivity score, sleep hours, study or work hours, social media hours, and daily screen time were the most influential variables. These findings suggest that digital behavior and lifestyle indicators may provide useful exploratory insights for stress-related analysis, although their predictive power remains limited. Therefore, the proposed approach is more suitable as an exploratory digital well-being assessment framework rather than a clinical diagnostic tool.]]>
Copyrights © 2026
