基于多准则的链式智能体遗传算法用于特征选择*

针对简单遗传算法用于特征选择精度不高、过早收敛的问题,提出了一种新的遗传算法——链式智能体遗传算法(LAGA),并与多准则(MC)相结合,从而提出了基于多准则竞争策略的链式智能体遗传算法(LAGA MC)用于特征选择。LAGA引入了链式智能体结构,智能体相互进行竞争选择和自适应交叉,自身进行自适应变异,从而使得该算法能够获得更精确的搜索结果;MC通过对基于单准则进行选择得到的特征子集进行特征位判断,从而确定出最终特征子集,以达到更全面的评价选择结果,获得识别率更稳定的特征子集。实验结果表明,LAGA搜索精度更高,LAGA MC获得的特征子集分类准确率更高、更稳定。

Feature selection with link like agent genetic algorithm combining multi criteria

According to low precision and over early convergence problems,the paper proposed a new genetic algorithm: link-like agent genetic algorithm(LAGA) and thereby proposed new method for feature selection with LAGA combining multi-criteria(MC).LAGA introduced link-like agent structure,competition selection,adaptive crossover and adaptive mutation,so it could obtain more precise search result.MC could judge the feature bits of the feature subset obtained through single criterion,thereby,obtained final feature subset to get more comprehensive and more stable result.The empirical results show that LAGA can get more precise search result;the feature subset obtained through LAGA MC has better classification rate and more stable classification result.