Study of Joint Optimization Algorithm for SVR Parameter and Feature Selection Based on Improved AGA

Given the significant impact of parameter tuning and feature selection on the prediction accuracy of Support Vector Regression (SVR), this paper presents a Joint Optimization Algorithm for SVR Parameters and Feature Selection based on the Improved Adaptive Genetic Algorithm (JOIAGA-SVR). The objective of this algorithm is to enhance predictive accuracy while eliminating redundant features that have a minor impact on the prediction target. To address the issues of traditional genetic algorithms being prone to getting stuck in local optima and having insufficient search capabilities in the later stages, we have improved the mutation conditions of the Adaptive Genetic Algorithm (AGA), adding discriminative conditions for individual mutation and refining the mutation probabilities of individuals, thereby enhancing the algorithm’s iterative ability in the later stages. To verify the proposed algorithm in this paper, comparative experiments were conducted on regression datasets from the UCI public database. Compared with the feature selection based on AGA and Genetic Algorithm (GA), as well as the joint optimization algorithm for SVR parameter and feature selection, JOIAGA- SVR has performed best in more than 90% of the datasets, and has eliminated redundant features without affecting predictive accuracy as much as possible. The experimental results demonstrate that the proposed algorithm is significantly more effective than existing methods in terms of feature selection and parameter optimization.