Improving Credit Scorecard Modeling Through Applying Text Analysis

In the credit card scoring and loans management, the prediction of the applicant’s future behavior is an important decision support tool and a key factor in reducing the risk of Loan Default. A lot of data mining and classification approaches have been developed for the credit scoring purpose. For the best of our knowledge, building a credit scorecard by analyzing the textual data in the application form has not been explored so far. This paper proposes a comprehensive credit scorecard model technique that improves credit scorecard modeling though employing textual data analysis. This study uses a sample of loan application forms of a financial institution providing loan services in Yemen, which represents a real-world situation of the credit scoring and loan management. The sample contains a set of Arabic textual data attributes defining the applicants. The credit scoring model based on the text mining pre-processing and logistic regression techniques is proposed and evaluated through a comparison with a group of credit scorecard modeling techniques that use only the numeric attributes in the application form. The results show that adding the textual attributes analysis achieves higher classification effectiveness and outperforms the other traditional numerical data analysis techniques.