AI-Driven Stress Detection: A Machine Learning Approach Using Physiological Signals

Stress detection is crucial for monitoring mental well-being, and machine learning algorithms offer an effective approach for classification. This study explores stress detection using physiological data from chest and wrist units, which exhibit an imbalance in data distribution. The dataset consists of labeled indices corresponding to different mental states: baseline (800,800 samples), stress (430,500 samples), amusement (253,400 samples), and meditation (537,599 samples). The data is split accordingly to train and evaluate multiple machine learning models. Several classification techniques, including Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), and Extreme Gradient Boosting (XGBoost), were applied. Additionally, Principal Component Analysis (PCA) was employed to reduce feature dimensions, followed by K-Nearest Neighbors (KNN) and XGBoost for enhanced classification. The performance of each model was assessed using accuracy, precision, recall, and F1-score. Results indicate that XGBoost outperforms other models with an accuracy of 98%, while PCA-based feature reduction combined with XGBoost achieves competitive results. LDA and QDA show moderate performance, with QDA achieving 89% accuracy. The study highlights the effectiveness of XGBoost in stress detection, demonstrating its potential for real-time mental health monitoring applications.