A method based on weighted F-score and SVM for feature selection

A novel feature selection method based on weighted F-score and SVM is proposed for the problems which inter-class overlapping and consistency of the features are ignored in traditional F-score feature selection method. Firstly, overlapping weight and consistency weight are introduced. Secondly, F-score value of every feature is calculated. Thirdly, F-score values of the features are ordered from high to low. Finally, the feature set of optimal recognition performance is selected by SVM. Simulation results for UCI machine learning database and experimental results for froth floatation plant data demonstrate that the proposed method can achieve better performance than traditional methods and has a good ability for generalization.