一种基于条件信息熵的多目标代价敏感属性约简算法的研究

代价敏感学习是数据挖掘和机器学习领域的重要课题.已有的研究方法多数针对单目标进行优化,并不适用于多目标代价敏感问题的解决.因此通过分析基于粗糙集领域的单目标代价敏感属性约简问题,定义了多目标代价敏感属性约简问题,并设计了一种简单高效的算法.在4个UCI数据集上的实验结果表明,该算法能获得令人满意的帕累托最优解集,以辅助用户进行方案的选择.

Multi- objective cost- sensitive attribute reduction based on conditional information entropy

Cost- sensitive learning is a focus in the field of data mining and the application of machine learning. The existing cost- sensitive learning researches usually focus on the algorithms dealing with a single- objective optimization rather than multi- objective cost- sensitive problems. This research defines and tackles the multi- objective attribute reduction problem with multiple types of cost. Experimental results on four UCI datasets indicate that this approach is effective to obtain satisfactory Pareto- optimal solution set and helpful to users in the scheme selection.