Optimizing Radial Basis Function Networks to Comprehend and Alleviate Perturbation Effects

This study primarily focuses on the design and enhancement of Radial Basis Function Networks (RBFNs). We’ve made these findings possible by creating two combined models: PSO-RBFN uses Particle Swarm Optimization (PSO), and GD-RBFN uses Gradient Descent (GD). We use the PSO and GD algorithms
separately to determine the optimal weights and biases for the RBFN. We select various algorithmic parameters to optimize the performance of each technique. We evaluate the performance of these hybrid models using various training and testing sets derived from time-series data. This study investigates the response of both models to training changes using different methods and subsequent testing. Experimental results demonstrate that the PSO-RBFN model outperforms the GD-RBFN model in terms of both accuracy and robustness. A paired t-test also demonstrates the statistical validation of PSO-RBFN over GD-RBFN.