基于CCS优化的FDT集成分类算法研究  

Research on FDT ensemble methods based on CCS optimization

作　　者：汪良楠 肖迪 

机构地区：[1]南京工业大学电气工程与控制科学学院,南京211816

出　　处：《计算机工程与应用》2018年第5期127-131,210,共6页Computer Engineering and Applications

基　　金：教育部高等学校博士学科点专项科研基金(No.20133221120012);国家自然科学基金(No.61403190)

摘　　要：模糊决策树在数据模糊化时,需要确定每个数量型属性的模糊语言项个数。另一方面,集成分类算法已成为提高模型准确率和稳定性的有效策略。提出了一种基于混沌布谷鸟(CCS)优化的FDT集成分类算法,首先用CCS算法确定数量型属性的模糊语言项个数,再通过bootstrap抽样生成FDT集成模型,最后采用OOB误差加权投票机制得到分类结果。通过4组UCI数据集验证,与其他分类算法对比,证明了该方法在分类精度上有明显的提升;同时,在处理缺失数据时,仍有较高的分类能力。When constructing the Fuzzy Decision Tree(FDT), data need to be fuzzy. The most important thing is to determine the fuzzy language item number for each quantitative attribute. On the other hand, ensemble classification algorithm has become an effective strategy to improve the model's accuracy and stability. So, a FDT ensemble classification algorithm based on Chaotic Cuckoo Search(CCS)optimization is proposed. Firstly, the model uses CCS algorithm to choose the fuzzy language item number of each quantitative attribute. Secondly, using the bootstrap sampling to generate FDT ensemble model, at last the model gets the classification results by OOB error weighted voting mechanism. Through 4 groups of UCI data sets experiments, it shows that compared with other classification algorithm, the method has significant improvement on classification accuracy. At the same time, in dealing with missing data, there is still a high classification ability.

关 键 词：模糊决策树 集成分类 混沌布谷鸟算法 投票机制 分类精度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]