A Comparison of Machine Learning Classifiers for Human Activity Recognition using Magnetic Induction-based Motion signals

Human activity recognition (HAR) is a growing research field with a wide range of applications. Magnetic induction-based human activity recognition system (MI-HAR) is a wearable-based HAR system proposed for capturing human motions and detecting activities based on the collected data. In this work, we focused on the performance analysis of different machine learning classifiers using synthetic magnetic induction based motion (MI-motion) signals. The main aim of this analysis is to compare the performances of six commonly used classifiers for HAR applications. Furthermore, we compared the classification performance using MI-motion data with the result reported in other studies using accelerometer data correspond to the same actions. Our results showed that Random Forest obtained the best performance of 91.5% on MI-motion data. Also, k-SVM and KNN models have respectively achieved accuracy of 91.4% and 86.4% on MI-motion data, which are both higher than the reported accuracy of 85.4% and 81.75 on accelerometer data.