Individual Optimal Feature Selection Based on Comprehensive Evaluation Indexs

Abstract

Individual optimal feature selection algorithm is simple and effective. Generally, single criterion is chosen for feature selection. But in this case, it is possible that features with good performance on other criteria will be neglected, and causing a negative impact to feature selection result. For the problem, the paper proposes a comprehensive evaluation model based on fuzzy correlation projection for evaluating the comprehensive indexs of individual features, and takes the results as the basis of feature selection. Definition to the model and its application procedure are described. Finally, the proposed method is applied to evaluate and select the underwater target recognition feature. The experimental result shows that the selected feature subset based on comprehensive indexs has a higher testing recognition rate than that based on single criterion, e.g. Relief-F indexs. Therefore, the proposed individual feature evaluation and selection method based on comprehensive indexs is feasible and effective.