Mukhtasir-Khail-Net: An Ultra-Efficient Convolutional Neural Network for Sports Activity Recognition with Wearable Inertial Sensors

The current prevalent approach of the Internet of Health and Medical Things entails proactively preventing disease onset through routine monitoring of individuals’ physical activities, making Human Activity Recognition (HAR) and Behaviour Analysis an important field of study. Sports Activity Recognition (SAcR) is a subset of HAR that focuses on identifying athletic movements. Computer Vision, Environmental Sensors, and Wearable Sensors are the three main methods for monitoring such activities. When the merits and limitations of these methods are considered, Wearable Sensors emerge as the most practical option. This study introduces the Mukhtasir-Khail-Net, a model with only 651 parameters that efficiently uses addition layers. When applied to six sports activities, the model achieves an impressive average accuracy of 98.865% using inertial sensor data generated by wearable sensors. For benchmark purposes, the presented model is also evaluated on the WISDM 11 dataset which is a Human Activity Recognition dataset from Fordham University and has achieved an accuracy of 94.24%.