Does Feature Reduction Help Improve the Classification Accuracy Rates? A Credit Scoring Case Using a German Data Set

Abstract

The paper broadly discusses the data reduction and data transformation issues which are important tasks in the knowledge discovery process and data mining. In general, these activities improve the performance of predictive models. In particular, the paper investigates the effect of feature reduction on classification accuracy rates. A preliminary computer simulation performed on a German data set drawn from the credit scoring context shows mixed results. The six models built on the data set with four independent features perform generally worse than the models created on the same data set with all 20 input features.