Reducing the Bank's Risk with Neural Network and K-Mean Algorithm

Banks engage in many problems regarding to facility granting. Low credit risk customers are preferred to provide them with facilities and other services. This research investigates the performance of k-mean and Neural Network for identifying high risk and low risk bank customers. We classify customers to low credit risk and high credit risk. Results of these two techniques are compared regarding to Basel Committee theory's ones. Results show that neural network model is more accurate and have more agreement with results of Basel Committee theory (94% of agreement) than clustering method (78% of agreement). Our contribution aspects in this research are as follows: (1) Considering new parameters (Totally 14 parameters) of clients, such as number of family members, amount of the loan and monthly income, for real customers; (2) Using the K-Mean algorithm in which normalized parameters are used as input in the neural network algorithm.