PPG Signal-Based Activity Recognition for Health and Fitness Monitoring

Photoplethysmography (PPG), commonly integrated into wearable devices, is widely utilized for detecting human activities and various physiological parameters. This study focuses on employing a feature-oriented and prime feature selection-based methodology in tandem with machine learning models like SVM and Decision Tree for effectively classifying human activities using wrist PPG data. The research underscores a clear relationship between essential features, relevant data points of features, their importance on target, and accurate classification of human activities derived from the PPG signals. The proposed approach demonstrates high accuracy, achieving 96.7% and 91.0% in two publicly available datasets. These results provide strong evidence of the efficiency and potential of features-based machine learning approaches in human activity classification from PPG signals.