Optimization of Bayesian Neural Networks using hybrid PSO and fuzzy logic approach for time series forecasting

Bayesian network is a form of graphical model for identification and calculation based on a group of influential variables, related to a probability distribution to deal with the complexity of the model. Providing flexible frameworks for the Neural Network training algorithm is one of the topics that has focused on many issues of the real world. On the other hand, Particle Swarm Optimization is a computational approach, an intelligent optimization, and the most popular algorithm that has been widely used for performing such types of optimization problems, which has faster convergence. Thus, in this paper, a hybrid innovative Gaussian Particle Swarm Optimization and fuzzy logic model is proposed to optimize the Bayesian Neural Networks in weight and structure. A hybrid algorithm to improve the performance of Neural Networks and the total number of weights in Neural Networks by determining the size of each particle, which is inspired by the prediction process. Furthermore, this method is used to forecast the economic growth based on time series dataset of Gross Domestic Product (GDP), which is one of the most widely used measures of an economy’s output. Then, this approach is evaluated and compared with time series techniques for modeling and forecasting selected data in terms of the performance of the standard and the correlation coefficient. The results show that using Bayesian Neural Networks improves structural models and minimizes operational errors. It combines different data types and prior knowledge, to avoid overfitting and to handle incomplete and noisy data, concerning to the quality of the learned network and the execution time. © The Author(s) 2025.