Explainable Machine Learning for Multi-Class Mental Stress Level Prediction: A Comparative Analysis of Ensemble and Conventional Models

Students’ and professionals’ mental stress has become one of the most serious public health issues; however, predictive models continue to be black boxes that do not get trusted by clinicians or teachers. In this research paper, we propose a comparison-based machine learning approach to solve the problem of classification of mental stress into three classes (Low, Moderate, High). Six machine learning algorithms such as Logistic Regression, K-Nearest Neighbors, Support Vector Machine, Random Forest, XGBoost, and LightGBM will be compared based on two feature spaces – all 20 features and the subset of 12 selected with the Mutual Information method. Evaluation was done based on the 1,100 samples survey dataset on psychological, physiological, social, environmental, and academic stressors. The best-performing single split was LightGBM, trained on features selected by the Mutual Information selection process, which resulted in an accuracy of 90.00%, Macro F1 score of 0.900, and Macro ROC-AUC of 0.987. It also showed the best mean cross-validation accuracy of 89.36% ± 0.84%. While tests show that the margin over Random Forest and Logistic Regression is directionally consistent but not statistically significant (as the number of samples is n = 5, therefore p > 0.05), it is statistically significant compared to XGBoost model (paired t-test, p = 0.011). To solve the problem of interpretability, SHapley Additive exPlanations (SHAP) were used in order to analyze the topperforming model. It showed that the key factors affecting students’ stress level are blood pressure, self esteem, and anxiety level, followed by strong contributions from academic performance and depression. The results show that compact and interpretable ensembles can compete with and even surpass the performance of complex black-box models and provide a good basis for practical applications.

Authors

Institutions

Publication Details

Journal
International Journal of Modern Science and Research Technology
Published
2026-11-19
DOI
https://doi.org/10.67317/ijmsrt26nov056
Primary Topic
Artificial Intelligence in Healthcare
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Explainable Machine Learning for Multi-Class Mental Stress Level Prediction: A Comparative Analysis of Ensemble and Conventional Models

Rajat Kumar Arya, Gourab Barui, Harjot Singh, Shivam Rana
International Journal of Modern Science and Research Technology
Artificial Intelligence in Healthcare
article

Explainable Machine Learning for Multi-Class Mental Stress Level Prediction: A Comparative Analysis of Ensemble and Conventional Models

Rajat Kumar Arya, Gourab Barui, Harjot Singh, Shivam Rana
article en

Abstract

Students’ and professionals’ mental stress has become one of the most serious public health issues; however, predictive models continue to be black boxes that do not get trusted by clinicians or teachers. In this research paper, we propose a comparison-based machine learning approach to solve the problem of classification of mental stress into three classes (Low, Moderate, High). Six machine learning algorithms such as Logistic Regression, K-Nearest Neighbors, Support Vector Machine, Random Forest, XGBoost, and LightGBM will be compared based on two feature spaces – all 20 features and the subset of 12 selected with the Mutual Information method. Evaluation was done based on the 1,100 samples survey dataset on psychological, physiological, social, environmental, and academic stressors. The best-performing single split was LightGBM, trained on features selected by the Mutual Information selection process, which resulted in an accuracy of 90.00%, Macro F1 score of 0.900, and Macro ROC-AUC of 0.987. It also showed the best mean cross-validation accuracy of 89.36% ± 0.84%. While tests show that the margin over Random Forest and Logistic Regression is directionally consistent but not statistically significant (as the number of samples is n = 5, therefore p > 0.05), it is statistically significant compared to XGBoost model (paired t-test, p = 0.011). To solve the problem of interpretability, SHapley Additive exPlanations (SHAP) were used in order to analyze the topperforming model. It showed that the key factors affecting students’ stress level are blood pressure, self esteem, and anxiety level, followed by strong contributions from academic performance and depression. The results show that compact and interpretable ensembles can compete with and even surpass the performance of complex black-box models and provide a good basis for practical applications.

International Journal of Modern Science and Research TechnologyVol. 4(11)
Amity University (IN)
Artificial Intelligence in Healthcare
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon