Hybridising feature subset selection with enhanced Deep Belief network for Human Activity recognition to Support Disabled Persons using internet of things–edge–cloud continuum

PURPOSE: Human activity recognition (HAR) utilises a wide array of sensors that produce vast amounts of data. Artificial intelligence (AI) is responsible for several HAR, which require significant computation and processing power. This article presents a new Hybrid Feature Selection Approach for Adaptive Human Activity Recognition to Support Disabled Persons (HFSA-HARSDP) approach in the IoT-Edge-Cloud continuum. The aim is to provide a promising strategy for next-generation smart healthcare and assistive IoT ecosystems via HAR assistance for disabled individuals. METHODS: The HFSA-HARSDP approach begins with data pre-processing, including outlier handling and normalisation. For dimensionality reduction, a hybrid approach combining minimum redundancy maximum relevance (mRMR) and recursive feature elimination (RFE) is employed to select the most discriminative features effectively. Finally, classification is performed using an enhanced deep belief network (EDBN) model to achieve accurate recognition of diverse human activities. RESULTS: The comparison analysis of the HFSA-HARSDP method showed superior accuracies of 99.34% and 99.27% on the HAR and WISDM datasets, respectively. CONCLUSIONS: The proposed model offers continual support of disabled persons via accurate detection of everyday activities in the IoT enabled healthcare environment.