Comparative Analysis of Artificial Hydrocarbon Networks versus Convolutional Neural Networks in Human Activity Recognition

Abstract

Human activity recognition (HAR) has gained interest in the research communities in order to know the behavior and context of users for medical, sports performance evaluation, ambient assisted living and security applications. Recent works suggest that convolutional neural networks (CNN) are very competitive machine learning techniques for HAR. Nevertheless, CNN require many computational resources, high number of parameter tuning, and many data samples for training. In this paper, we present a comparative analysis of a novel technique, artificial hydrocarbon networks (AHN), with CNN on human activity recognition classification task. We choose to compare AHN with CNN given that it is a very well-suited machine learning technique for HAR. We show that AHN architecture is simpler to set up than CNN, it needs less hyper-parameter configuration and has a slightly better accuracy performance.