Implementing high accuracy and model reliability in credit scoring using improved EWOA-BP algorithm

Credit assessment is a key problem in the field of finance, which can predict whether a user has the risk of delinquency, thereby reducing the loss of bad debts. BP neural network has been widely used in credit evaluation because of its excellent ability of data learning and induction. However, it also has disadvantages, such as slow convergence and being susceptible to local outliers. Swarm-based intelligence algorithms offer advantages such as simplicity of use, rapid convergence, and powerful global optimization capabilities, making them effective optimization algorithms. In order to improve the convergence speed and prediction accuracy of BP neural network in credit evaluation, this paper constructs EWOA-BP model based on improved whale algorithm. Firstly, we propose a multi-strategy Enhanced Whale Algorithm (EWOA) by introducing a new non-linear decrease approach based on cosine function, a unique exploration technique that employs leader-based adaptive tangent travel, and a novel exploitation strategy, which can effectively balance the exploration and exploitation, reduce the risk of local optima trapping, and accelerate the convergence speed while ensuring the accuracy. Secondly, the Enhanced whale algorithm is used to optimize the neural network and construct the EWOA-BP model. Finally, the Enhanced Whale Algorithm (EWOA) is verified and analyzed by using 23 classical benchmark functions, and the performance is excellent. The EWOA-BP model is validated on three credit assessment datasets, and by comparing it with 10 contemporary algorithms, the results show that the EWOA-BP model obtains better performance in personal credit assessment, and comprehensively concludes that EWOA-BP algorithm is effective and more competitive.