The Effect of Feature Selection on Classification Algorithms in Credit Approval

Data mining applications usually use datasets with large dimensions. Unfortunately, the large dimension in a dataset affects the processing time and outcome of the classification. One solution to this problem is to select relevant features for the reduce dimensions. The selection of the right features can increase the accuracy of the classification process. This study proposes a proper feature selection model for increasing the accuracy for specific classifier models by comparing several existing feature selection models and some of the classifier. The feature selection, we use information gain, gain ratio, and correlation-based feature selection (CBFS) while the classifier we employee K-Nearest Neighbor (K-NN), Support Vector Machine (SVM), Artificial Neural Network and Decision Tree. Evaluation is done using several public databases from UCI Machine Learning Repository. Evaluation results indicate that all feature selection methods are increasing the accuracy, but Decision Tree only increases in CBFS. Based on the result, feature selection does not always improve classifier accuracy but depends on the characteristics and algorithms used.