Deep Learning-Based Motion Activity Recognition Using Smartphone Sensors

Abstract

MHealth systems establish a new way to transfer the health service to remote places. These systems offer significant benefits for continuous health monitoring. Motion activity recognition is one of the challenging mHealth use cases that incorporates continuous data collection and analysis of measurements. The main goal of this research is to analyze physical activity data. We employ measurements from the WISDM lab dataset 1 . These data are collected from participants performing motion activities. This data is then used by deep learning algorithms to predict special activities. In particular, CNN and CNN-LSTM algorithms are used to compare their accuracy, which resulted in approximately 95% and 97% respectively. Thus, the CNN-LSTM has higher accuracy in this analysis.