<![CDATA[The increasingly intensive use of social media in everyday life has various implications for the psychological condition of users, one of which is an increase in stress levels due to high usage duration and excessive exposure to information. This condition necessitates an analytical approach to understand and predict user stress levels more objectively. This study aims to compare the performance of Random Forest, XGBoost, and CatBoost algorithms in classifying the stress levels of social media users based on digital behavior data, as well as to identify behavioral factors that contribute to these stress levels. This study uses a quantitative approach based on data mining with a dataset consisting of 667 social media user data. The research stages include data collection, preprocessing, modeling using three machine learning algorithms, and model performance evaluation using accuracy, precision, recall, and F1-score metrics, reinforced with confusion matrix and feature importance analysis. The results show that Random Forest produced the best performance with an accuracy value of 0.84, precision of 0.86, recall of 0.83, and F1-score of 0.84, followed by CatBoost with an accuracy of 0.80 and XGBoost with 0.78. Feature importance analysis shows that Daily Screen Time and Happiness Index are the most influential variables in determining the stress level of users, which aligns with previous findings in digital mental health research.]]>
Copyrights © 2026
