一种改进的自适应GA-SVM参数选择研究

支持向量机是一种学习机器,决定SVM性能的因素是核函数的选取,但其参数的选择大多是依靠经验,一般不能获得最佳函数逼近效果,一定程度上限制了该算法的发展。将改进的自适应遗传算法与支持向量机相结合,设计了一种自动优选支持向量机模型参数的方法。该方法根据适应度值自动调整交叉概率和变异概率,减少了遗传算法的收敛时间并且提高了遗传算法的精度,确保了SVM参数选择的准确性。将该方法应用于脱机手写汉字的识别,结果表明由该方法所得的SVM具有较好的泛化能力。

Parameter Selection of Support Vector Machines Based on an Improved Adaptive Genetic Algorithm

The selection of kernel function is a decisive factor on the performance of support vector machines （SVM）. However, most users select parameters by rule of thumb, and frequently fail to generate the optimal parameters effect for the function, which has restricted the effective use of SVM to a certain extent. This paper pres- ents an automatic parameter selection method for SVM combining the adaptive genetic algorithm. This method selects crossover probability and mutation probability according to the fitness values of the object function, and thus reduces the convergence time and improves the precision of GA, insuring the accuracy of parameter selection. The application of this method in off-line handwriting Chinese character recognition demonstrates an improvement of the generalization performance for support vector machines.