One-D Convolution Neural Network Models for Human Activity Recognition using mHealth Datasets
K M Yogesh, S. Arpitha, Ibrahim Gad · 2023
The task of identifying a person's specific physical activity based on sensor data is a complex problem in time series classification. It falls under the field of research called "Human Activity Recognition (HAR)." Successfully training a machine learning model for this task often requires significant domain knowledge and the application of signal processing techniques to extract useful features from the raw data. Deep learning methods such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) have proven to be effective, as they can automatically extract features from the raw sensor data and achieve state-of-the-art results. In this study, we propose a multi-label One-Dimensional Convolutional Neural Network (1D CNN) for detecting human physical activity in the mHealth dataset. Our Deep Learning (DL) model demonstrates promising results, achieving an accuracy of 99.58% in categorizing various physical activities. Among the different approaches for feature extraction and movement prediction from raw sensor data, Convolutional Neural Networks (CNNs) are found to be the most effective method.