Performance Analysis of Adaptive Filter and Machine Learning Algorithms for Heart Rate Estimation Using PPG Signal

Abstract

Photoplethysmography (PPG) signal provide advanced and simple ways for estimating heart rate (HR) information as an unremarkable system on wearable devices. In this paper, we analyze the performance of adaptive filter and machine learning (ML) algorithms for estimation of HR during physical activity. Three cascades recursive least square (RLS) and cascades normalized least mean square (NLMS) adaptive filters are developed and combined using convex combination scheme to reduce motion artifacts (MA) from the recorded PPG signal. Then, ML based spectral tracking algorithms is applied, to locate the spectral peak corresponding to HR. Four different supervised ML algorithms (Support Vector Machine, Decision Tree, K- Nearest Neighbor and Logistic Regression) are examined to track the spectral peaks and the decision tree out performs all three algorithms with an accuracy of 98.96%. Experimental results on the PPG datasets including 23 subjects used in the 2015 IEEE signal processing cup showed that the proposed approach has a very good performance by achieving an average absolute error (AAE) of 1.98 beats per minute (BPM) and the personal correlation coefficient of 0.9899. AAE result proved that the proposed method provides accurate HR estimation performance in comparison with other existing works.