A Modified Neural Network Classifier with Adaptive Weight Update and GA-Based Feature Subset Selection
This paper proposes a new neural network classifier system with adaptive weight update. The system is divided into two sections namely, feature subset selection section and classification section. Genetic algorithm is introduced to complete feature subset selection to save the cost of training dataset. Classification section is inspired by a further research on the weight coefficient of membership function in “Data-Core-Based Fuzzy Min-Max Neural Network”(DCFMN).The modified classifier can improve the classification accuracy when training data is much smaller than testing data where this situation often occurs in real word due to its capacity of updating its weight coefficient while testing data online. This ability is really indispensible to classify unlabeled dataset such as field data for fault detection. The proposed modified classifier is tested on data-base available online. Results demonstrate the good qualities of this new neural network classifier.
KeywordsDCFMN genetic algorithm feature subset selection Fuzzy min-max neural network Classifier Weight value update
Unable to display preview. Download preview PDF.
- 2.Bezdek, J.C., Pal, S.K.: Fuzzy Models for Pattern Recognition, Piscataway, NewYork (1992)Google Scholar
- 10.Harrag, A., Saigaa, D., Boukharouba, K., Drif, M., Bouchelaghem, A.: GA-based Feature Subset Selection Application to Arabic Speaker Recognition System. In: 11th International Conference on Hybrid Intelligent Systems (HIS), pp. 382–387. IEEE Press, New York (2011)Google Scholar