Evolving trends in smart home activity recognition with edge computing and ethical best practices

Abstract

Human Activity Recognition (HAR) in smart homes is a vital breakthrough in constructing intelligent systems for monitoring Activities of Daily Living (ADL). It improves healthcare, security, and minimize energy requirements. This study utilizes ARAS data to construct and test sophisticated predictive models for multiresident activity systems based on state-of-the-art machine learning classifiers, ensemble approaches, and deep neural network structures. We utilize advanced feature extraction and selection methods such as Information Gain, Recursive Feature Elimination (RFE), and Random Forest Importance to balance model performance and computational efficiency. Our work involves incorporating edge computing paradigms with ethical frameworks to respond to privacy issues and real-time processing needs. The hybrid architecture we put forth showcases improved performance with accuracy of 99.6% and 99.8% for households A and B respectively, while being low in latency and energy consumption. Additionally, we present thorough ethical guidelines and privacy-preserving methods to promote ethical deployment of HAR systems in home environments. Experimental verification on a variety of scenarios ensures the scalability and robustness of our method, qualifying it as a potential solution for next-generation smart home systems.