Enhancing Credit Scoring: A Hybrid Feature Selection Algorithm for Loan Approval

Since the financial crisis in 2008, the crucial rule of credit scoring in bank credit risk management has been determined. Banks require efficient credit scoring models to distinct creditworthy customers. For making these models, banks usually use the existence data of their customers. They need to select suitable features for making these models, and unnecessary features with no significant effects are required to be removed. Since, there are many submitted features for each customer, for selecting the best feature selection model, lots of approaches must be checked; therefore, choosing essential features is a time-consuming approach, and it is leading to a NP-Hard problem. In this research, a feature selection model based on profit in the credit ranking of bank customers has been proposed. The proposed model is based on the hybrid algorithm Particle Swarm Optimization and Simulated Annealing. In this model, a precise classification of creditworthy customers is implemented; and the costs of each selected feature are applied. The PSO-SA hybrid method was first used to determine features, and then a series of individual and combined classification algorithms are applied to the dataset. The results are compared based on the advantage of each classifier and its accuracy. In this research, the credit data of Tose'e bank is acquired. The outcomes showed that in the classification approaches, individual algorithms like, Artificial Neural Network, and among the combined algorithms, the AdaBoost has the highest accuracy and efficiency. Furthermore, to show the high degree of accuracy, Min–Max and Z-Score techniques are compared.