A Clustering-Based Gauss Chaotic Mapping Particle Swarm Optimization for Auto Labeling in Human Activity Recognition

Human activity recognition (HAR) has become a vital research field owing to its significant contribution to pervasive computing. In order to train practical models based on machine learning in HAR, a large number of samples need to collect sensor information and mark each sample as belonging to a specific activity. Therefore, designing an efficient labeling method is one of the important research topics in HAR. In this study, we applied an effective clustering method based on Gauss chaotic mapping particle swarm optimization (GCMPSO) to improve the accuracy of automatic labeling results. Six human activities with 561 features were used to evaluate the GCMPSO clustering method, including walking, sitting, standing, walking upstairs, walking downstairs, and lying in repose. The results of these six activities were automatically labeled by K-means and GCMPSO clustering methods. Our results show that the automatic labeling performance of the GCMPSO clustering method in HAR is superior to the K-means algorithm.