Accelerometer based classification of elbow flexion and extension exercises
Gamze Uslu, Şebnem Baydere, Ata Tekin, Feryal Subaşı · 2020
In this study, we propose a human activity recognition (HAR) system for monitoring the adherence of the patients to the elbow flexion and extension physiotherapy exercise routines in their daily environments. The proposed solution utilizes single wrist-worn accelerometer data with a one-class classification (OCC) approach where training data are collected for the target class only and data that represent all other classes are produced artificially from the target activity data. We analyzed the system with four different classifiers; KNN (k-nearest-neighbors), SVM (Support Vector Machines), logistic regression and Naive Bayes classifiers. The results reveal that SVM achieves a median success rate of 100% and 83,3% in detection of target and other classes respectively in intra-subject evaluation as well as reaching a median success rate of 90% and 100% in inter-subject case on a per-subject basis, emerging as the best performing classifier among the ones we study. Our one-class approach is also observed to outperform one-class SVM of LIBSVM, which severely suffers from parameter selection issue.