基于懒惰学习的显露模式分类  

Lazy Learning-based Classification by Emerging Patterns

作　　者：田卫东[1] 温勇[1] 

机构地区：[1]合肥工业大学计算机与信息学院数据挖掘与智能计算实验室,合肥230009

出　　处：《小型微型计算机系统》2016年第4期753-757,共5页Journal of Chinese Computer Systems

基　　金：国家"八六三"高技术研究发展计划项目(2012AA011005)资助;国家自然科(61273292)资助

摘　　要：现有基于显露模式的分类方法主要通过精简显露模式的数量以构建实用的轻量级分类器,然而对显露模式集的过度精简会损害数据信息的完整性,进而影响分类器性能.本文提出LLEP分类器,采用懒惰学习策略,将分类器的构建推迟到分类阶段进行,以在获知待分类事务信息的基础上,构建出更具针对性的局部分类器;对于显露模式的冗余消除问题,采用了等价类方法来快速划分包含重复信息的显露模式,以保留鲁棒性更优的显露模式参与分类.本文在UCI机器学习库27个数据集上的实验表明,LLEP分类器同11经典种分类器相比,在分类准确度上表现出了良好的性能.Classifier constructing scheme widely applied in dominant emerging pattern classification methods is to as far as possible reduce emerging patterns for constructing an effective light classifier.But too much reduction maybe affects the integrity of data information,and finally decreases these classifiers＇ performance.In this paper a newemerging pattern classification method called LLEP,which is Lazy Learning Emerging Pattern,is proposed.Based on lazy learning strategy,LLEP postpones the constructing of classifier from training step to classifying step so that LLEP could collect information from the data to be classified and then could integrate this information for constructing more specific classifier.M eanwhile for eliminating redundant emerging patterns,equal-class method is applied to quickly divide emerging patterns into different groups according to information they reflected,and on the basis of these groups,a set of more robust emerging pattern could be chosen for classification.Experiments on 27 datasets showthat LLEP outperforms other 11 classical known classifiers.

关 键 词：显露模式 等价类 懒惰学习 覆盖率 鲁棒性 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]