The banking system tend to internalize scoring according to Basel II & III and their Central Bank regulations. Consequently, these banking systems are in dire need of credit scoring models. In this study, first, we present a probabilistic neural network (PNN) algorithm for credit scoring of bank customers optimized by means of a genetic algorithm. Based on data from legal customers of one Iranian bank, its performance is compared with seven common machine-learning algorithms. Then we developed a new hybrid performance metric, called probabilities of credit scoring correctness, by combining several performance metrics. The banking system has proposed several credit-scoring models. Models such as single classifiers, hybrid models, and ensemble models determine the class of customers (good or bad). In order to calculate the expected loss and unexpected loss, banks need the probability of default. In general, the proposed model can utilize m performance metrics and n classifiers; the larger m and n, the more reliable the customer class estimates will be. In fact, the purpose of this paper is to create a hybrid approach for credit scoring Iranian banks' clients, thus obtaining the probability of default and credit risk models for the banking system, especially the weak banking system.