A relative evaluation of the performance of ensemble learning in credit scoring

Credit scoring prediction is a focus of banking sector to identify trickery customers and to reduce illegal activities. The usage of ensemble classifiers in machine learning plays a vital role in prediction problems. The aim of this study is to analyze the accuracy of the ensemble methods in classifying the customers as good risk group or bad risk group. In this paper experiments are conducted using three ensemble methods namely AdaBoost, Bagging, Random Forest combined with three learning algorithms. Feature selection is applied for selecting important attributes from credit card dataset. This paper provides an assessment on performance of the ensemble classifiers taken for this study.

I. Introduction

Data mining is the process to uncover useful information's which are unknown from a large dataset [1]. It contains many techniques like clustering, classification and Association rules. There are many classification algorithms like Support vector machine, neural network, desicion tree, and naive bayes. Support Vector Machine constructs a hyper plane in high dimensional space that maximizes the margin between two classes [2]. Neural networks learn by example and then classify the instances. Decision tree is used in this study as a base learning algorithm. Naive bayes is more suitable for high dimensional data. Each algorithm has its own pros and cons. Prediction for a classification problem cannot be obtained effectively by using any one of the classification algorithms. Instead of depending on one single classifier, we can combine two or more base classifiers using ensemble method and the accuracy of classification can also be improved.