Abstract: For original Bacterial Foraging Algorithm（BFA） exists the problems of slow convergence and large amount of calculation, firstly, by improving bacterial population size, movement step length and introducing the iterative termination conditions to improve the original BFA, and then applies it to Support Vector Machine（SVM） parameters optimization. Based on Iris standard test data sets, experiments take the penalty factor [C] and kernel parameter [γ]of Gaussian kernel SVM as the optimization objects, and analyze the optimization performance of genetic algorithm, particle swarm algorithm, original and improved BFA, the results verify that applied the improved bacterial foraging algorithm in SVM parameter optimization has superiority.

Key words: Bacterial Foraging Algorithm（BFA）, parameters optimization, Support Vector Machine（SVM）

摘要： 针对原有细菌觅食算法收敛速度慢、计算量大的问题，首先通过改进细菌种群大小、细菌运动步长、引进迭代终止条件改进原有细菌觅食算法，然后将其应用到支持向量机的参数优化上。实验以Iris标准测试数据集为依托，以高斯核支持向量机中核参数[γ]和惩罚因子[C]为优化对象，分析了遗传算法、粒子群算法、原有的和改进后的细菌觅食算法的寻优性能，验证了将改进后的细菌觅食算法应用到支持向量机参数选择上具有优越性。

关键词: 细菌觅食算法（BFA）, 参数优化, 支持向量机（SVM）