A Discriminative Probability-Based Weighted Algorithm for Human Action Recognition

Traditional machine learning classification algorithms, such as DT, KNN, RF, and SVM, because of the curse of dimensionality, their accuracies are often limited in recognizing human actions. In this paper, we first propose a discriminative probability-based weighted algorithm, which uses discriminative probability in both offline and online phases and weights them to obtain a more reliable recognition result. Secondly, we collect large amount of human action Channel State Information(CSI) data samples using commercial WiFi devices. Experimental results show that regardless of which model we choose from DT, KNN, RF, and SVM, its recognition accuracy is significantly improved after combining with our proposed algorithm, with a maximum improvement of 8.2%.