基于动态调整的GA-SVM多分类二叉树的方法

基于遗传算法的支持向量机决策树多分类方法仍然存在错误累积的问题,累积的错误往往使分类准确率下降,分类效果变差,存在全局优化缺陷的问题;并且在每个节点进行二分类常用的实数编码切分方法,效率低下。针对这两方面的问题,提出从根节点开始逐层构造二叉树,运用二进制编码的遗传算法进行每一阶段的二叉树构造。二进制编码对于每个节点的分类以及进行交叉、变异更高效,不用考虑从什么位置切分。针对越靠近根节点产生的误差对后续节点分类误差的累积影响,提出一种动态调整的方法,此方法对每个节点赋予权重再进行对权重的调整使得整体的分类误差减小,最终得到二叉树的全局最优,从而提高分类精度。通过实验并进行五折交叉验证表明,DABT-SVM比多种传统的支持向量机多分类算法在全局优化能力和分类精度上有很大的提升。验证了该算法的可行性和有效性。

GA-SVM Binary Tree Multi-Classification algorithm Based On Dynamic Adjustment

The problem of error accumulation still exists in SVM binary tree multi-classification which based on genetic algorithm. Error accumulation invariably result in low classification accuracy rate and defection in global optimization. The use of real number coding segmentation in every node causes inefficiency. Regarding the problems mentioned above,this paper states the idea of gradually building up binary tree structure from root node and using binary coded genetic algorithm to structure binary tree for each stage.Binary coding increases the efficiency in crossover, mutation and segmentation.Furthermore,In order to solve the problem accumulation errors,we propose a method of dynamically adjusting.This method gives a weight to each node then make adjustments on nodes so that the overall classification errors decrease.Experiment results show that the proposed method has greatly improved on the global optimization capability and classification accuracy.