Comparative analysis of feature selection techniques for EEG-based stress detection using a hybrid CNN-LSTM model

Stress has an impact no yet not just on physiological health but also cognitive performance, making its detection in time enough to provide preventive care important. Electrophysiology, such as electroencephalography (EEG), will provide an effective and non-invasive method of monitoring neural responses that accompany processing stress. Here we describe an optimized deep learning framework that incorporates Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) architecture for EEG-based stress recognition. The model proposes a hybrid feature selection method with combined Archimedes Optimization Algorithm (AOA) and Analytic Hierarchy Process (AHP) features in order to optimize efficiency and discriminative power. The CNN-LSTM-AOA-AHP model is evaluated against traditional classifiers including Classification Tree, Ensemble(ES), K-Nearest Neighbor (KNN), Support Vector Machine (SVM) and variants. In the experiments, we demonstrate that the classification accuracy was between 46-55% for the models with the traditional classifiers, while CNN-LSTM-AOA-AHP achieved 96.1% accuracy and demonstrated better performance of precision, recall and F1 score. Findings confirmed the robustness and reliability of optimization-based deep learning in capturing spatial and temporal EEG features and provided a solid baseline to inform future work in the development of realtime mental stress detection systems.