Breaking New Ground in HAR with Enhanced Weighted k-NN Algorithm

In recent times, the rise in demand for IoT-based Human Activity Recognition (HAR) applications across diverse sectors such as health monitoring, elderly care, gait analysis, security, and Industry 5.0, has been noteworthy. A critical challenge encountered in these developments is the accuracy of reference models, prompting this research to focus on enhancing model precision through advanced machine learning (ML) methodologies. The study meticulously analyzed a distinct machine learning strategy, represented in the k-Nearest Neighbor (k-NN) approach. Data was meticulously gathered from 102 individuals, aged between 18 and 43, and segmented into training and testing datasets. These datasets were instrumental in the supervised learning phase, utilizing refined ML techniques. This rigorous process enabled the accurate identification of twelve daily activities, ranging from sedentary behaviors like sitting and laying to dynamic movements such as walking, jogging, and cycling. The findings revealed that Weighted k-Nearest Neighbor (Wk-NN) approach outperformed with a remarkable accuracy of 98.6%. This study conclusively demonstrates that the enhanced ML approaches significantly boost the accuracy of daily activity classification, with enhanced Wk-NN approach leading slightly in performance metrics.