<![CDATA[Stress among university students negatively impacts their academic progress and mental health. Early detection is crucial for targeted intervention. This research designs and evaluates a machine learning model using the XGBoost algorithm to predict stress among undergraduate students of Information Technology Education at Universitas Brawijaya. Utilizing the CRISP-DM methodology, the study processes data from the DASS-42 questionnaire and academic records. The workflow included data pre-processing to handle missing values and class imbalance, followed by model training and evaluation. Six scenarios tested prediction targets (five-level multi-class and binary ‘Normal’ or ‘Stress’), data handling, and hyperparameter tuning. Results indicate that the binary classification model was significantly superior. The best model, utilizing original data and default parameters, achieved an accuracy of 97.87%. Evaluation proved its reliability, achieving 100% recall for the ‘Stress’ class, ensuring no at-risk cases were missed (0 False Negatives). Feature importance analysis identified Mother’s Education as the dominant predictor. The research output includes a functional dashboard prototype equipped with LIME interpretation for individual case analysis.]]>
Copyrights © 2026
