HUMAN ACTIVITY RECOGNITION: A DUAL PERSPECTIVE ON DEEP LEARNING AND HEALTHCARE MONITORING

Abstract

In recent years, Human Activity Recognition (HAR) has become a significant research field in the healthcare monitoring area as it involves the analysis and interpretation of various human behavioral patterns by employing wearable sensors, smartphones, Internet of Things devices and intelligent computational algorithms. The recent breakthroughs in deep learning have resulted in the development of HAR applications ranging from elderly care to rehabilitation monitoring, assisted living and remote patient supervision. But current HAR systems have a number of limitations, such as distinguishing similar activities is challenging, handling subtle differences in the behavior is difficult, limited anomaly detection ability and lack of interpretability in a healthcare setting. This research gives a thorough survey about deep learning-based HAR techniques and introduces a new HAR framework for healthcare applications that is enhanced by incorporating change-point detection and anomaly analysis mechanisms. The proposed framework is based on Density-Based Spatial Gradient Aligned Clustering of Applications with Noise for pattern identification, Gated Recurrent Temporal Echo Memory Unit for activity recognition, Generative Cross Window Temporal Attention Adversarial Network for change-point detection, Memory-Augmented Hidden Information Bayesian Markov Model for uncertainty estimation, and Semantic Distance Regularized Autoencoder for anomalous activity detection. The framework is tested on the MHEALTH dataset following the pre processing, dimension reduction and multi-domain feature extraction. The experimental results showed that the proposed model had the activity recognition accuracy of 98.91%, precision of 98.92%, recall of 99.06% and F-measure of 98.99%. The proposed autoencoder model had lower reconstruction errors and better clustering quality for anomaly detection. The proposed framework has been successfully applied to healthcare-related HAR applications such as monitoring of elderly, rehabilitation support, remote healthcare supervision, and smart assisted living system.