The Research Of SVM-based Feature Selection And Its Ensemble Method

With the rapid development of the computer technology and network technology,the emergence of large data sets make people obtain the amount of information which is increasing at an unprecedented speed,at the same time,which also lead to irrelevant or redundant data emerge continually.Therefore, how to obtain useful information quickly is becoming people concerns.At present, the feature selection in data mining is a technique to eliminate data noise and as a common means to preprocess the data according to a criterion selected the optimal combination from the original features, which has been became a hot research field of pattern recognition.The paper using the new technologies of feature selection based on support vector machine which has good generalization and ensemble classifiers. Based on support vector machine, the main research contents are feature selection and the means of ensemble classifiers, the research object is the factors which influence the ability of support vector machine, research regulatization parameter C and kenel function parameter deeply and through ensemable various classifiers to futher improve the generalization in data mining.The main duties include the following two aspects:Firstly, RGS algorithm is designed. The ReliefF algorithm provides a priori information to GA, the parameters of the SVM mixed into the gene encoding, and then using genetic algorithm finds the optimal feature subset and support vector machine parameter combination. Finally, experimental results show the effectiveness of the proposed algorithm.Secondly,an ensemble classification algorithm RGSE is presented from the difference of the classifier.This algorithm is based on combination of RGS classifiers.The algorithm is mainly by reducing the correlation of classifiers to enhance its ability of ensemble classification,and the experiments show that the thinking of this algorithm is better than Bagging and Boosting.