A modified hybrid feature-selection method based on a filter and wrapper approach for credit risk forecasting

Guotai Chi and Mohamed Abdelaziz Mandour

Abstract

Sifting through features to find the most related ones is known as feature selection. This paper introduces a feature-selection technique based on a modified approach in order to improve classification performance using fewer features: the chi-squared with recursive feature elimination (χ²-RFE) method. It combines χ² as a filter approach with recursive feature elimination as a wrapper approach. The algorithm developed for the χ²-RFE method is superior to six other algorithms in measures of average performance, with acceptable computing time. This is demonstrated by application to a data set of Chinese listed companies with a sample size of 47 172 and 535 characteristics, and the efficacy of the χ²-RFE algorithm is further confirmed by an experiment on a German data set with a sample size of 1000 and 24 characteristics. Since it can be challenging to achieve high accuracy and good performance in measures related to imbalanced data with only a few features, we extensively analyze the potential of our modified feature-selection framework, χ²-RFE, to provide a solution.