Streaming feature selection method via adaptive neighborhood
Presently,though various types of streaming feature selection techniques have been developed,the implementations of most of those methods require enough domain knowledge before parameter setting.To solve this problem,a streaming feature selection method via adaptive neighborhood is proposed.Firstly,a new neighborhood relation is defined,which calculates the radius based the natural distribution of samples in each category,so that the neighborhood can be constructed adaptively.Secondly,the correlation and redundancy of streaming features are analyzed by using the neighborhood-based dependency.Finally,through using a general process of streaming feature selection,it is not difficult to seek out a satisfied feature subset.To verify the effectiveness of the proposed algorithm,a comparative analysis was carried out over 18 datasets with three advanced streaming feature selection methods.The experimental results demonstrate that the streaming feature selection results generated by the proposed method can significantly improve the average classification accuracy of the test samples over both K-nearest neighbor(KNN)and support vector machine(SVM)classifiers with the superiority of more than 5.68% .
万方数据
维普
