Artificial Neural Network approach to credit risk assessment

With the wide use of credit facilities utilized by the corporate customers, the accuracy of predicting credit worthiness is of utmost importance to the financial institutions. This paper reports on a novel approach to credit risk assessment which incorporates the impact of financial, industry and economic factors specific to a particular industry, using artificial neural network technology. For the prototype, Credentials, the hotel industry of Sri Lanka was selected. The input parameters consider growth rates on a quarterly basis to determine the gearing position. Two neural networks are being trained for cashflow growth and debt growth prediction respectively using backpropagation learning rule. The predicted gearing growth rate for the next four consecutive quarters is generated as the output. The modular design incorporates Web services to export hotel and economic data and Hebbian based classification. According to the evaluation, the best accuracy levels were obtained for Colombo region five star hotels, where cashflow neural network and debt neural network yielded 75.0% and 62.5% success rates respectively. Overall, the evaluator panel was satisfied with the concept, approach, technical aspects, prototype and the reporting facilities of Credentials.