Abstract

The academic stress, anxiety and the uncertainty as regards the career matters have enormous psychological effects on the students who are at this important juncture of education. The prior support system and targeted guidance with the proper mechanisms of early detection are required. The machine Learning algorithms help facilitate in practical deployment and targeted guidance for students at their age group. The most underappreciated mental health challenges in modern education are mental stress among high school students. In this research Shannon Entropy based feature selection has been used. Through that the three ensemble classifiers namely Random Forest, XG Boost and Cat Boost were used to predict mental stress levels from structured questionnaire. The data was collected in five psychological domains from 500 respondents answering 15 questions through five-point Likert scale. The XG Boost delivered commendable 91% accuracy and AUC = 0.996, while random forest followed by 95% of accuracy and AUC of 0.998. CatBoost achieved the highest test accuracy (97%) with AUC of 1.000, though five-fold cross-validation yielded a more conservative estimate of 92.8% (±3.31%). The high test-set performance reflects the cluster-derived target variable methodology; external clinical validation remains essential. In this research the most discriminating features are of motivation related questions, Q14 and Q 15. The framework is small enough to be deployed in practice and comprehensible enough to be used in the context of target student support programs.

Statistics