Enhanced Stress Detection via Heart Rate Data: A Feature Selection and Stacking Ensemble Approach

Stress is a major issue that affects both the personal and the professional and social environments. There are several methods for detecting stress, but most are unreliable and have a large number of false negatives. The goal of this study is to construct a machine learning-based system to accurately identify stress level based on heart rate data. This work focuses on feature selection and using a stacking ensemble model to improve accuracy, reduce misclassification, and increase reliability. As stress becomes a critical issue for individual well-being and organizational productivity, stress detection systems need to become more effective. A data-driven approach is more consistent and easily expandable, while the traditional approaches are unreliable. This study acknowledged the stress detection problem by providing a reliable and efficient method. The method of detecting stress includes preprocessing the heart rate data, handling outliers, evaluating importance of features using the ANOVA test, selecting relevant features using K-Best method, and normalizing data. Multiple state-of-the-art machine learning models such as Random Forest, KNN, Decision Trees, and MLP were trained, and a stacked ensemble model combining their outputs was used for improved prediction. The results indicate that the stacking model presents the best performance, obtaining an accuracy score of 99.95%, with zero false positives and false negatives. This model requires longer running time, as compared to other models, it yields better accuracy and reliability. The stacking model proposed in this study shows promise as a stress detection tool, but rigorous validation by professionals is required before it is applied in real-world scenarios to monitor and control stress levels.