基于改进遗传算法的确定范围内SVM参数选择

SVM Parameters Selection Based on Improved Genetic Algorithm Within Determined Range

提出了在确定范围内采用改进遗传算法自动选择支持向量机(SVM)参数的方法。首先,通过分析SVM的参数取值与模型性能的关系,确定各参数的限定范围,然后,分析了遗传算法的不足,通过引入递阶编码、小生境共享和自适应交叉概率等技术对其进行改进,最后,以所确定的范围作为搜索区间,利用改进的遗传算法自动选择SVM参数。实验结果验证了该方法的有效性。

An automatic selection method of Support Vector Machine （SVM） parameters is proposed based on an improved Genetic Algorithm （GA） within the determined range. First,the limited range for each parameter is derived by analyzing the relation between SVM parameters value and model capability,and then the technologies including hierarchical encoding,niches share and adaptive crossover probability are introduced to overcome the shortcomings of standard GA. Finally,the determined range can be regarded as a search region of SVM parameters and the improved GA is used to automatically search the optimal parameter values of SVM. The experiments are given to illustrate the effectiveness of the method.