ABSTRACT This research usesself-organizing maps (SOM) in order improve the ability of the pattern recognition techniques including neural networks and K-nearest neighbour used to forecast the credit risk of borrowers from Bank of Agriculture (BOA) Sokoto. In this work, a hybrid approach to building the credit scoring model was proposed using the unsupervised learning based on self-organizing map (SOM) to specifically improve the discriminant capabilities of K-nearest neighbour and Neural networks. Within the two-stage scheme, the knowledge (i.e., prototypes of clusters) found by SOM is considered as input to the subsequent pattern recognition models. The results from BOA, Sokoto data indicate that the two-stage models improved the performances of Neural Network and K-nearest neighbour from 96.3% and 95.7% to 97.3% and 96.3% respectively.