Gaussian Mixture Support Vector Machines for Credit Scoring

Abstract

Credit scoring assesses the risk associated with lending to individuals or institutions and is of great importance. To do the scoring, classification methods, based on machine learning, classify the individuals into creditworthy and uncreditworthy. We utilize both application scoring and behavioral scoring for the classification of different datasets. We apply a novel preprocessing method and also a novel denoising method based on the Gaussian Mixture Model and then combine it with the Support Vector Machines classifier. Our method is specifically useful to treat imbalanced datasets, and our results show that a well-treated balanced dataset improves the standard models significantly. We compare our approach with the standard benchmarks and show a higher overall accuracy.

Keywords: classification, machine learning, preprocessing, denoising, imbalanced datasets