1* Research Scholar, Faculty of Engineering, Design and Automation, GNA University, Phagwara, Punjab, India
2 Associate Professor, School of Engineering, Design and Automation, GNA University, Phagwara, Punjab, India
3
Received: 12th Dec, 2025 | Revised: 12th Feb, 2026 | Accepted: 11th March, 2026 | Available Online: 29th March, 2026
Mental stress has been a major issue of concern since it impacts on the way individuals think, feel and operate in their day- to-day lives. With the increase in the availability of psychological and behavioral data researchers have begun to apply data-based approaches to gain improved insights into stress in a manner that can be extended to larger groupings. Nonetheless, practice has demonstrated that the application of a single predictive model tends to produce disproportionate results. A model can work well given a particular set of conditions but one cannot say the same about the intricate combination of the academic pressure, emotional condition, lifestyle habits, and social issues that can contribute to stress. In order to solve this problem, the current research examines an ensemble learning framework of stress prediction. Rather than relying on a single algorithm there is a combination of several classifiers so that each has its own advantages and disadvantages compensated. The core aim of this strategy is to have more reliable predictions. The secondary goal is to compare this combined strategy and separate machine learning models. The framework is put to the test with the help of Student Stress Factors dataset provided in Kaggle, which covers such information as academic workload, psychological condition, lifestyle behavior, and social influences. A number of classification methods is used individually and then combined by the majority vote. This enables the end prediction to be a collective decision and not an individual decision. The experimental findings indicate that the performance of individual models varies widely with the values of the accuracy varying between approximately 79 percent and 96 percent. Comparatively, the ensemble technique yields more stable results that yield an accuracy of 97.89 percent and an average value of high precision, recall, and F1-score. Further Stratifiedcross-validation outcomes also suggest that the model is reliable in other data splits.
Keywords: Mental Stress Prediction, Machine Learning, Ensemble Learning, Classification Models, Performance Analysis.
How to cite this article: Gupta P, Pandit A, An Ensemble Machine Learning Framework for Mental Stress Prediction and Performance Evaluation..Int J Drug Deliv Technol. 2026;16(3): 478-489. DOI: 10.25258/ijddt.16.1.55
Source of support: Nil.
Conflict of interest: None.