A novel feature learning for image classification using wrapper approach in GA

In many pattern recognition applications, limiting the number of features is a very important requirement due to high dimensional cost as well as the risk of “overfitting” imposed by the high-dimensional feature vectors. Feature subset selection addresses the dimensionality reduction problem by determining a subset of available features which is most essential for classification. A novel feature learning for image classification is proposed here using wrapper approach in Genetic Algorithm. The proposed method applies GA for feature subset selection and neural network for classification. The method operates by trying to choose the subset of features which lead to the largest margin of class separation between two classes. Experiments are conducted on four benchmark datasets of iris, seed, glass and wine and then used on one domain dataset of rice. Comparison of the proposed approach is made with other approaches like Multi-SVM and GA-LDA to demonstrate its effectiveness and efficiency. Analysis of the experimental results shows that the proposed method outperforms the other two approaches in classification accuracy.