CorLMI-FSA: An Efficient Feature Selection Approach for Stress Classification Using Physiological Signals

Mental stress in academic settings significantly impacts students' health and performance. Physiological signals such as Electrodermal Activity (EDA) and Heart Rate Variability (HRV) offer valuable insights for stress classification. However, the multidomain nature of these signals often introduces redundancy and irrelevant data, reducing both classification accuracy and computational efficiency [1]. Hence, this paper proposes an efficient feature selection algorithm, CorLMI-FSA (Correlation-Logistic Mutual Information Feature Selection Algorithm), which dynamically adjusts redundancy weights through a logistic function to capture nonlinear dependencies among features. The proposed algorithm achieves accuracies of 96.82% and 95.84% for binary and three-level stress classification tasks using Gradient Boosting (GB), enhancing computational efficiency by selecting only the most relevant features. These results underscore CorLMI-FSA's potential as a low-complexity solution for realtime stress monitoring in wearable devices, facilitating timely intervention.