A Novel Hybrid Classification Model For the Loan Repayment Capability Prediction System

Abstract

Classification is a powerful tool in Data mining to predict the loan repayment capability of a banking customer. This paper evaluates the performance of various classification algorithms and selects the most appropriate one for predicting the class label of the credit data set as good or bad. Feature selection is a data pre-processing technique refers to the process of identifying the most beneficial features for a given task, while avoiding the noisy, irrelevant and redundant features of the dataset. These irrelevant noisy features results in a poor accuracy for the selected classifier. In order to improve the accuracy of a classifier, the feature selection plays a vital role as a data preprocessing step. Feature selection technique reduces the dimensionality of the feature set of the dataset. This paper has two objectives. First objective is to find out the best classifier algorithm for the credit data set using two different tools such as weka and R. Here the experiment proved that Random Forest performs better for loan repayment credibility prediction system. The second objective is to evaluate the performance of various feature selection methods based on Random Forest classification method. Also a novel hybrid model is developed for the same.