An Automated Hybrid Higher Order Neural Classifier

摘要

Inspired from human brain perception, considering correlative relations of diverse input features facilitates pattern classification. It is usually addressed to Higher Order Neural Networks (HONNs) which involve superior specifications to traditional neural networks. In this paper, we propose a novel Genetic Algorithm based Hybrid Higher Order Neural Classifier (GA-HHONC) for handling classification problems. Two main experiments over original data sets and data sets with selected features are carried out. In the case of original datasets, it shows improvements about 3.5% and 0.6% in compare with the best accuracy reported in Chen & Shie (2008) for classifying the Balance Scale and Iris data sets, respectively. In addition, by using same method for feature selection, it has been shown that proposed method perform more accurate than methods presented in Shie & Chen (2008). Improvements are about 1.7%, 1.3% and 0.2% in classification of Pima, Iris and Breast cancer data sets, respectively.