Estimating Nonlinear Regression Parameters using Particle Swarm Optimization and Genetic Algorithm

Abstract

Obtaining parameter estimates for nonlinear regression model using gauss-newton and gradient-based methods present some complex analytical challenges. In this paper we investigated the effectiveness and simplicity of Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) on five nonlinear regression models with varying level of complexities. We developed function in R-programming for each models and performed 30 independent runs for at least 100 iterations for both PSO and GA. We evaluated PSO and GA performance in view of computation time, residual error produced and compared our results with values published online. Based on the results obtained, PSO significantly outperform GA in view of computation time and quality of parameter estimates. Even so, GA required fewer iterations and produced fairly accurate results. Further investigation shows that PSO and GA are both competitive, effective, simple to implement, and can be considered reliable for obtaining the parameter estimates of different nonlinear regression tasks.