Enhanced Human Activity Recognition (E-HAR): Leveraging Sensor Fusion, Placement and Algorithmic Strategies for Improved Activity Recognition

Abstract

Activity recognition, a crucial aspect of healthcare monitoring, relies on accurate data processing from various sensors for effective analysis. This paper proposes a framework Enhanced Human Activity Recognition (E-HAR) to optimize activity recognition systems by integrating sensor fusion techniques and algorithmic selection strategies. Leveraging diverse datasets encompassing multiple sensor types and placements, our study explores the performance of various algorithms across distinct sensor data categories. The framework E-HAR prioritizes Dataset D3, characterized by consistent high performance across algorithms, establishing it as a reliable source for activity recognition model training. Decision Tree (DT) and Multi-Layer Perceptron (MLP) algorithms emerge as versatile choices due to their robustness across datasets. Furthermore, sensor type and placement significantly impact recognition accuracy. Vitals and ankle sensors demonstrate superior performance, emphasizing their efficacy in achieving higher F1 scores. The combination of these sensors showcases the potential for enhanced accuracy through sensor fusion. By outlining an optimal pathway for activity recognition, this research contributes a structured approach for healthcare practitioners and researchers to effectively design and implement activity recognition systems, enhancing the reliability and accuracy of healthcare monitoring in diverse contexts.