文献类型：会议论文

英文题名：Multi-Relational Classification in Imbalanced Domains

作者：Xu, Guangmei[1];Bao, Hong[1];Meng, Xianyu[2]

第一作者：徐光美

通讯作者：Xu, GM[1]

机构：[1]Beijing Union Univ, Inst Informat Technol, Beijing 100101, Peoples R China;[2]Liaoning Univ Technol, Sch Comp Sci & Engn, Jinzhou 121001, Peoples R China

第一机构：北京联合大学智慧城市学院

通讯机构：[1]corresponding author), Beijing Union Univ, Inst Informat Technol, Beijing 100101, Peoples R China.|[1141734]北京联合大学智慧城市学院;[11417]北京联合大学;

会议论文集：3rd International Conference on Intelligence Computation and Applications

会议日期：DEC 19-21, 2008

会议地点：China Univ Geosci, Wuhan, PEOPLES R CHINA

主办单位：China Univ Geosci

语种：英文

外文关键词：imbalanced dataset; multi-relational data mining (MRDM); sampling; mutual information; naive Bayes

摘要：This paper discusses the problem of multi-relational classification on imbalanced datasets. To solve the class imbalance problem, a new multi-relational Naive Bayesian classifier named R-NB is proposed. the attribute filter criterion based on mutual information is upgraded to deal with multi-relational data directly and the basic sampling methods include under-sampling and over-sampling are adopted to eliminate or minimize rarity by altering the distribution of relational examples. Experiments show, with the help of attribute filter method, R-NB can get better results than those without that. And, experiments show that multi-relational classifiers with under-sampling methods can provide more accurate results than that with over-sampling methods considering the ROC Curve.