A New Approach for Attribute Importance Measure Based on TCA-SSVM
Yanfeng Fan, Dexian Zhang, Huacan He · 2009
The lack of heuristic information is the fundamental reason that affects the attribute selection in data mining. Spatial hypersurface plays a very important role in the classification problem which reflects the characteristic of class attribute and condition attributes. In this paper, a new measure for determining the importance level of the attributes based on partial derivative distribution of the output corresponding to the inputs is presented. For more convenience, we present a new SVM model called TCA-SSVM which could use simple algorithm to solve the optimization problem to acquire the classification hypersurface. The proposed approach is experimentally evaluated in two datasets and the results prove that it can improve the validity of the problem and lead to interesting results.