Bankruptcy Prediction in Banks by Fuzzy Rule Based Classifier

In this paper, a fuzzy 'if-then' rule based classifier is employed to predict bankruptcy in banks. This classification problem is formulated as a multi objective combinatorial optimization problem, where the rule base size is minimized and classification rate is maximized. Modified threshold accepting is applied to solve this optimization problem. The efficacy of the classifier is tested on the well-known US banks bankruptcy data set. It performed very well and in the case of 2 partitions outperformed the multi layer perceptron by yielding higher average classification rate and lower average Type-I error.