A Machine Learning Framework for Stress Detection Using Photoplethysmography (PPG) Signals

Stress detection using machine learning has gained significant consideration due to its potential applications in mental health monitoring and well-being assessment. This study presents an efficient framework for stress classification, incorporating preprocessing, feature extraction, and model training techniques. Initially, raw signal data undergoes preprocessing, including bandpass filtering and segmentation, to enhance data quality. Features are extracted and labeled, followed by the dataset being split into training and testing sets. Multiple machine learning models, including K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Naive Bayes, and AdaBoost, are evaluated for classification, with the Gradient Boosting Classifier (GBC) as the proposed model. Experimental results demonstrate that GBC outperforms other classifiers, achieving an accuracy of 99.77%, precision of 99.76%, recall of 99.76%, F11-score of 99.77%, mean squared error (MSE) of 0.0046, and root mean squared error (RMSE) of 0.0962. These results demonstrate how well our method works to differentiate between stressed and non-stressed states, providing a viable way to identify stress in real time.