Building classification models for customer credit scoring

Abstract

Credit scoring is a quantitative method. It has been used in the banking industry. The major challenge of credit scoring is essentially to classify credit customers into different risk groups. As known methods, Hidden Markov Model (HMM) have been proposed as a classification technique. It can be utilized to build a classification model. The developed process consists to distinguish preciously the profitable customers from bankrupt customers using all possible characteristics describing the applicant. The important and the difficult task in credit scoring problems arises when a banker decide whether to grant or not grant a loan. In this order, the aim of this work is to investigate the performance of HMM for building a good classifier for credit scoring. In this paper, we also provide a set of classification models to score customer by combining the HMM and Baum-Welch procedure. There are two phases in this model: firstly, we train the initial model by HMM techniques estimations, secondly we use the iterative procedure Baum-Welch to generate HMMs parameters and building new models. Experimental results show that the proposed model can build the highest accuracy classifier for credit scoring datasets. The implementation of HMM with Baum-Welch model allows lenders and bankers to develop techniques to measure customer credit risk.