Banks use the credit score to rank potential individuals among loan customers. This score helps the banks to determine the high risk customers among loan customers. The scoring model incorporates essential data from personal customer data, credit histories and customer behavior. This research proposes a combined scoring model based on two data mining techniques, clustering analysis and classification, called “Cluster Analysis of multi-predictors (CLAMP).” This combined strategy constructs the model in two phases. The first phase is a clustering process which uses the X-mean clustering algorithm. This process will partition training data into k groups where k is determined by information measure criteria. The second phase is classifier selection from J48 (decision tree model), Naïve Bayes (probability model), logistic regression (statistical model) and multi-layer perceptron (artificial intelligent model). The criteria for selecting classifier are based on partitioning data into 40% training set for building a model, 30% validation set for selecting the best classifier within a group and 30% test set to reject overfitting model. The classifier using CLAMP shows a better accuracy than the classifier alone.
