Fig. 3.

Feature selection. Feature selection methods including filtering, wrapper, and embedded. a The filtering methods rank the features according to a certain characteristic or correlation, and specify a threshold value or directly select the top ranked features. b The embedded method adopts the way that the feature is directly selected by the model. The model obtains the weight coefficient of each feature after trainings and selects the best feature according to the coefficient. c The wrapper methods take model performance as a criterion to judge the quality of features or feature subsets, and gradually retain or remove several features