Designing Wireless Sensor Network Data Based Machine Learning Approach for Accurate Human Activity Recognition

Human activity recognition (HAR) using a Wireless Sensor Network (WSN) is essential for understanding human behavior and predicting future actions. With HAR, smartphones, and classification models are important aspects. This study explores the use of WSNs for classifying human activities by analyzing time series data captured using wireless devices. This paper aims to design an intelligent model for classifying human activity from WSN data. This paper combines multivariate dynamic mode decomposition (MDMD), Hankel Block (HB), and an ensemble classifier for classifying physical human activity. The MDMD decomposes time series data into principal modes, from which features are extracted for analysis and classification. The Hankel matrix enhances the estimation of hidden oscillatory modes when integrated with MDMD. Ensemble classifiers are employed to classify the extracted feature. The proposed model’s performance is evaluated on three datasets, UCI-HAR, Opportunity, and WISDM. The obtained results proved that the proposed model is superior to competing models, and it can be deployed in WSN to support different real-time applications.