关键词: 人工智能, 特征选择, 包装方法, 过滤器方法, ReliefF算法

Abstract: Feature selection is one of the most important issues in system modeling, data mining and pattern recognition. A new hybrid feature selection method, named RRK, is proposed which is based on the wrapper and filter methods. This method includes two steps: (1) feature preselection phase, in this step an improved algorithm named NReliefF based on ReliefF and a dimension reduction algorithm named DR are given to finish featureweight and useless feature reduction; (2) feature selection phase, in this step the result of the feature preselection is used as the input, and the NReliefF and KNN algorithms are combined to select the optimal feature which is based on the highest accuracy rate. Both practical data sets of occupant type and standard data sets of Imagine in UCI are used to verify the proposed feature selection method. Results show that both the accuracy rate and efficiency of the classification are significantly improved by this method.

Key words: artificial intelligence, feature selection, wrapper method, filter method, ReliefF algorithm