摘要
传统的三支决策模型是依靠专家经验来设置阈值的,从而阻碍了三支决策模型在许多领域的广泛应用。针对此不足,提出不需要依赖于专家经验的基于网格搜索的最优阈值生成算法,即以三支决策风险损失函数为模型,以决策风险最小为目标,以网格搜索为手段,以样本的条件概率为搜索空间,找出能使风险损失最小的参数组合——最优阈值。最后将以该算法得到的阈值构建的三支分类器与贝叶斯分类器分别应用于UCI部分数据集,结果显示三支分类器分类性能更优,从而说明该算法有效。
The traditional three-decision-making model relies on the experience of experts to set the threshold, thus impeding the wide application of three-decision-making model in many fields. To minimize the decision-making risk, a computational model of the risk-loss was built, and a new classification algorithm which needs no priori knowledge was given. The algorithm used model conditions to determine the range of parameters value which minimized the risk-loss, then divided the range into several equal grids, got the smallest range of parameters through searching these grids, and the smallest range was the optimal threshold. At last, a three-decision-making classifier was built by using the threshold, and then this classifier and Bayesian classifier were applied to part of UCI data sets. The comparison shows that the performance of three-decision- making classifier is superior, which shows the effectiveness of the algorithm.
出处
《计算机应用》
CSCD
北大核心
2012年第8期2212-2215,共4页
journal of Computer Applications
基金
国家自然科学基金资助项目(61175027
11071271)
关键词
三支决策
最优阈值
网格搜索
风险损失
数据集
three-decision-making
optimal threshold
gid searching
risk-loss
data set
