Towards Accurate Stress Classification: Combining Advanced Feature Selection and Deep Learning

Stress is a significant health issue that affect both physical and mental health. The application of deep learning and machine learning techniques for physiological signal processing has become popular in detecting and classifying stress. In this research, we developed a fresh approach to classify stress on multiple levels using advanced machine and deep learning models. The PhysioNet SRA database was used to collect physiological signals which were pre-processed using a pipeline of filters and normalisation techniques to remove noise and artefacts. These signals were subjected to various feature extraction algorithms to determine the most pertinent characteristics for categorisation. On the acquired features, dimensionality reduction techniques were applied in order to further improve classification performance. The evaluation of Random Forest model outscored all the other classic ML models, achieving an accuracy of 93.77% and 82.42% in classifying stress on two levels and three levels respectively. Among the DL models, the LSTM model is most effective, achieving an accuracy of 95.67% and 88.70% in classifying stress on two levels and three levels respectively. The proposed approach provides a more comprehensive analysis of stress levels, which can help individuals manage stress more effectively.