结合互信息的多目标属性约简

Algorithm of attributes reduction based on multi-objective evolutionary and mutual information

为了获得决策系统中更好的相对属性约简,提出一种基于互信息的多目标属性约简算法。该算法首先根据互信息寻找核属性集;然后以最小属性子集和最大互信息为目标,定义新的适应度函数,在粒子运动方程、克隆及自适应变异的共同作用下进化;并通过非支配排序及精英保留策略寻找满足目标的Pareto最优解。通过UCI标准数据集上的对比测试结果表明,算法能够有效地对决策系统进行约简。

In order to obtain better relative attribute reduction in decision systems,this paper proposed an algorithm of attri-butes reduction based on multi-objective evolutionary and mutual information.The algorithm first searched attribute core set by mutual information,then aiming at the least reduction of attributes sets and maximum mutual information,designed a new adaptive function,redefine motion equation and adaptive mutation operation,searched Pareto optimal solutions based on non-do-minated sorting and elitism strategy.Experimental results with UCI data sets show that the algorithm can effectively reduce the decision system and obtain ideal reduction results.