基于自适应隶属度函数特征选择.pdfVIP

Vol. 32, No. 4 ACTA AUTOMATICA SINICA July, 2006 Feature Selection Based on Adaptive Fuzzy Membership Functions1) XIE Yan-Tao1 SANG Nong1 ZHANG Tian-Xu2 1(Institute for Pattern Recognition and Artificial Intelligence, Huazhong University of Science and Technology, Wuhan 430074) 2 (Key Laboratory of Ministry of Education for Image Processing and Intelligent Control, Huazhong University of Science and Technology, Wuhan 430074) (E-mail: nsang@hust.edu.cn) Abstract Neuro-fuzzy (NF) networks are adaptive fuzzy inference systems (FIS) and have been applied to feature selection by some researchers. However, their rule number will grow exponentially as the data dimension increases. On the other hand, feature selection algorithms with artificial neural networks (ANN) usually require normalization of input data, which will probably change some characteristics of original data that are important for classification. To overcome the problems mentioned above, this paper combines the fuzzification layer of the neuro-fuzzy system with the multi-layer perceptron (MLP) to form a new artificial neural network. Furthermore, fuzzification strategy and feature measurement based on membership space are proposed for feature selection. Finally, experiments with both natural and artificial data are carried out to compare with other methods, and the results approve the validity of the algorithm. Key words Membership function, feature selection, arc