A Modified Parameter Optimization Method of Support Vector Machine Based on Genetic Algorithm

Given the computational demands and the less-than-ideal efficiency of genetic algorithm-driven support vector machine parameter optimization methods, and taking into account the distribution of classification accuracy within the parameter space, this paper proposes to shift the search space of the genetic algorithm from the conventional real number space to a discrete parameter space. This adjustment significantly reduces the search space of the genetic algorithm. Experiments conducted on eight standard datasets demonstrate that this method not only achieves almost the same classification accuracy but also significantly enhances the genetic algorithm's speed in identifying the optimal support vector machine parameters. As a result, the computational load is reduced by approximately 50 times when compared to the traditional genetic algorithm and by approximately 10 times when compared to the grid search algorithm.