I. Introduction
In recent years, wearable devices have seen significant improvements in their sensor technology, enabling better perception ability. This has made human activity recognition more accurate, and its now widely used in various real-world scenarios. For instance, it is used for gesture recognition in human-computer interaction [1], posture estimation in health detection [2], and the monitoring violent actions in crime prevention [3]. Because of its immense application value, human activity recognition has garnered great attention. Wearable sensors come in many types, including kinematic sensors like acceleration sensors, gyroscopes, and magnetometers. To achieve precise activity recognition, researchers often analyze multiple sensor data [4].