Abstract

Improving Credit Scoring Performance using Two-Stage Technique

Ibrahim Anas and Sam Olu Olagunju

---

In this study an unsupervised learning based on Self-Organizing Map (SOM) was used to specifically improve the discriminant capabilities of Classification And Regression Trees (CART) and Artificial Neural Networks (ANN) used to predict the credit risk of borrowers from Bank of Agriculture (BOA) Sokoto. In this work, a two-stage approach to building the credit scoring model was proposed using SOM and CART. Within the two-stage scheme, the knowledge (i.e., prototypes of clusters) found by SOM were considered as input to the subsequent classification model (i. e. CART). The results from BOA, Sokoto data indicate that the two-stage model improved the performances CART from 96.3% to 96.7%. This therefore suggests that the integration of SOM algorithm into CART made SOM+CART hybrid model to outperform the stand-alone CART model. Keywords: Credit Scoring, Self-Organizing Map (SOM), Classification and Regression Tree (CART), Creditworthy, Non-Creditworthy.

PDF