Health Gaming based Activity Recognition Using Body-Worn Sensors via Artificial Neural Network

Integrating body-worn sensors with virtual reality technology has revolutionized user engagement in health gaming and e-learning applications. This paper presents an activity recognition system using the Wireless sensor data mining dataset, designed to enhance interaction in Virtual reality environments. The system employs a multi-step process that includes signal preprocessing with a second-order Butterworth filter, feature extraction via Fourier transform, Random forest, and Power spectral density, followed by feature selection using Recursive feature elimination. Classification is performed using artificial neural network to ensure accurate and real-time activity recognition. The proposed approach addresses challenges in computational complexity and sensor data noise by integrating optimized signal processing techniques with advanced machine learning models. The integration of activity recognition with Arduino and Unity platforms demonstrates significant improvements in user interaction quality within Virtual reality based health and educational gaming. Experimental results demonstrate that our proposed method achieves 78% accuracy in activity recognition, outperforming Random Forest (75.9%) and SVM-Radial (65.77%), resulting in a 2.1 percentage point improvement. Additionally, the integration of optimized preprocessing, feature selection, and an artificial neural network classifier enhances computational efficiency, leading to a noticeable reduction in processing latency, thereby ensuring real-time performance suitable for Virtual Reality applications.