Life Expectancy Prediction Based on Particle Swarm Optimization Support Vector Machine Regression

Abstract

This study delves into the application of Particle Swarm Optimization (PSO) algorithms in the parameter optimization of Support Vector Regression (SVR) models to enhance the accuracy of predicted average life expectancy. An adaptive PSO algorithm augmented with random perturbations is introduced. Through a fitness-function-guided dynamic adjustment mechanism, this method circumvents the premature convergence issue typical of conventional PSO, locating the optimal penalty parameter C and kernel parameter for the SVR model. These parameters are then incorporated into the SVR model for training and regression prediction, achieving optimal data fitting and generalization effects. Comparative analysis reveals that the PSO-enhanced SVR model outperforms the traditional SVR across various performance metrics. Notably, there is a marked improvement in the coefficient of determination, a significant re-duction in root mean square error, and mean absolute percentage error, affirming the advancements of the improved PSO-SVR model in prediction accuracy and reliability.