A Novel Human Activity Recognition Using Spiking Neural Network
Huy Hoang Nguyen, Tuan Khoi Nghiem, Mai Phuong Hoang, Duc Minh Nguyen · 2024
This research is primarily focused on Human Activity Recognition (HAR). HAR aims to forecast a user's daily activities by analyzing time series data collected from wearable sensors. However, contemporary behavioral datasets often overlook an essential aspect of phone listening activity, which is a central focus of this research. Additionally, most existing research in behavioral classification has not considered energy optimization through Artificial Neural Network (ANN) models. To address these gaps, this study introduces a dataset capturing human responses to phone calls and proposes a behavior classification model employing Spiking Neural Networks (SNN s) to address these issues. The proposed solution is multifaceted, encompassing measures such as model accuracy, parameter count, and energy consumption. Our proposed model achieved outstanding performance, boasting an accuracy of 96.72% on our dataset and remarkable energy consumption of approximately 7.42×10-08 J/inf. Moreover, experimental results on two renowned datasets, SHAR and VCI HAR, affirm the method's efficacy, demonstrating its applicability beyond our specific dataset.