一种基于NNIA多目标优化的代价敏感决策树构建方法

摘要

本文提出了一种基于非支配邻域免疫算法(NNIA,Nondominated Neighbor Immune Algorithm)多目标优化的代价敏感决策树构建方法.将平均误分类代价和平均测试代价作为两个优化目标,然后利用NNIA对决策树进行优化,最终获取了一组Pareto最优的决策树.对多个测试集的测试结果表明,与C4.5算法和CSDB(Cost Sensitive DecisionTree)算法比较,本文方法不仅在平均误分类代价和平均测试代价两方面均可以取得优于两者的性能,而且获得的决策树具有更小的规模,泛化能力更强.%A novel method of constructing the cost-sensitive decision trees based on multi-objective optimization is proposed in this paper. The average misclassification cost and the average test cost are treated as the two optimization objectives. NNIA (Non dominated Neighbor Immune Algorithm) is exploited to optimize the decision trees. And some Pareto decision trees are finally ob tained. Experimental results show that.compared with the C4.5 algorithm and CSDB (Cost Sensitive Decision Tree) algorithm, the proposed method in this paper can not only outperform these two methods in terms of the two above objectives but also achieve smaller size of the decision trees and stronger generalization ability.