In data mining, selecting the critical feature subset from original features is very important. Feature selection cannot only reduce the dimensionalities of datasets but also improve the performance of the classifier on the selected data. In this paper, a hybrid feature selection (HFS) method based on multiple feature selection methods is proposed, it is composed of four single feature selection methods: CFS, ReliefF, CFS-GA and LVF. For proving the improvements of feature selection methods in classification, five medical UCI datasets are employed as the studied materials. Based on the critical features filtered from the medical datasets, HFS is compared with the four single feature selection methods on three testing classifiers of AdaBoostM1, MultiBoostAB and BayesNet, and the results show that HFS almost improves the classification efficacy of the classifiers better than the single feature selection methods in medical data classification.
展开▼
