Abstract

Due to the rapid expansion of the Internet of Things (IoT), supportive systems for healthcare have made significant advancements in both diagnosis and treatment processes. To provide optimal support in clinical settings and daily activities, these systems must accurately detect human movements. Real-time gait analysis plays a crucial role in developing advanced supportive systems. While machine learning and deep learning algorithms have significantly improved gait detection accuracy, many existing models primarily focus on enhancing detection accuracy, often neglecting computational overhead, which can affect real-time applicability. This paper proposes a novel hybrid combination of Sparse Gate Recurrent Units (SGRUs) and Devil Feared Feed Forward Networks (DFFFN) to effectively recognize human activities based on gait data. These data are gathered through Wearable Internet of Things (WIoT) devices. The SGRU and DFFFN networks extract spatio-temporal features for classification, enabling accurate gait recognition. Moreover, Explainable Artificial Intelligence (EAI) assesses the interoperability, scalability, and reliability of the proposed hybrid deep learning framework. Extensive experiments were conducted on real-time datasets and benchmark datasets, including WHU-Gait and OU-ISIR, to validate the algorithm’s efficacy against existing hybrid methods. SHAP models were also employed to evaluate feature importance and predict the degree of interoperability and robustness. The experimental results show that the method, combining Sparse GRUs and Tasmanian Devil Optimization (TDO)-inspired classifiers, achieves superior accuracy and computational efficiency compared to existing models. Tested on real-time and benchmark datasets, the model demonstrates significant potential for real-time healthcare applications, with an AUC of 0.988 on real-time data. These findings suggest that the approach offers practical benefits for improving gait recognition in clinical settings.