Evaluation of Feature Selection Methods for Predictive Modeling Using Neural Networks in Credits Scoring

摘要

A credit-risk evaluation decision involves processing huge volumes of raw data, and hence requires powerful data miningtools. Several techniques that were developed in machine learning have been used for financial credit-risk evaluationdecisions. Data mining is the process of finding patterns and relations in large databases. Neural Networks are one of thepopular tools for building predictive models in data mining. The major drawback of neural network is the curse ofdimensionality which requires optimal feature subset. Feature selection is an important topic of research in data mining.Feature selection is the problem of choosing a small subset of features that optimally is necessary and sufficient to describethe target concept. In this research an attempt has been made to investigate the preprocessing framework for feature selectionin credit scoring using neural network. Feature selection techniques like best first search, info gain etc. methods have beenevaluated for the effectiveness of the classification of the risk groups on publicly available data sets. In particular, German,Australian, and Japanese credit rating data sets have been used for evaluation. The results have been conclusive about theeffectiveness of feature selection for neural networks and validate the hypothesis of the research.