Abstract:

 A new feature extraction method combined with depth function was intended to solve the problem of the traditional feature extraction method of which the efficiency of the feature extraction method was not very satisfactory when the dataset had many different distributions. After studying the Euclidean distance,the relationships of all features according to the kernelized spatial depth function in order to get the all data features and extract similar features were judged to diagnose the same class or the outlier class. Finally, three different standard UCI databases running on three different dimensions and three different running times were simulated in order to demonstrate the algorithm’s efficiency. The result showed more adaptive features. The results showed that the kernelized spatial depth function for feature extraction achieved a satisfactory result and could propose a new idea for the Relief algorithm.

Key words: kernelized function, Relief algorithm, depth function, feature extraction, Euclidean distance;   Mahalanobis distance; , Gauss kernel function, statistical depth