Hierarchical classification and visualization with multiple feature ranking criteria

Hierarchical classification is consistent with human cognitive thinking mode and easy to understand. In the process of classification, the feature ranking method has a great impact on the classification result and convergence rate. In this paper, a hierarchical classification method with multiple feature ranking criteria is proposed, including Least Number of Overlapping Interval Samples (LNOIS) method, Maximum Average Distance (MAD) method, Minimum OTSU-MSE (MOTSU-MSE) method, etc. The proposed method is intuitive and concise without adjusting the specific super parameter. To enhance the interpretability of this method, a visual system is designed based on JavaScript programming language. The method is applied to the recognition of human daily behavior, and effective features are extracted and filtered according to the characteristics of signals. The hierarchical classification model is trained based on OTSU-MSE method, and 93.06% F1 score is obtained.