摘要
针对单分类提出了一种集成金字塔模型(Single-Class Classification Integrated Pyramid Model,SCCIPM),由综合获取层、辅助判定层、核心分类层和结果优化层4个独立协同的层面组成。其中综合获取层由改进的KNN以及优化的1-DNF两个分类方法组成,主要用来获取反例,其结果均提交到辅助判定层投票后得到可靠反例;核心分类层通过多次迭代更新可靠反例,每次迭代建立一个分类器;结果优化层根据核心分类层所建立的不同分类器优化选择最终分类器。仿真实验表明,当正例占整个样本的50%以下时,SCCIPM较其它方法优势明显,在解决单分类上具有良好的分类性能。
A Integrated Pyramid Model is proposed for Single-Class Classification（Single-Class Classification Integrated Pyramid Model,SCCIPM）.This model is composed of four independent collaborative layers by Comprehensive obtain layer,Assistant judgment layer,Kernel classification layer and Optimization layer.Comprehensive obtain layer formed by improved KNN and optimized 1-DNF two classification methods,mainly for obtaining negative examples,results are submitted to Assistant judgment layer voting reliable negative examples.Kernel classification layer update reliable negative examples by several iteration,each iteration establish a classifier.Optimization layer according to Kernel Classification layer of different classifiers,optimally select the final classifier.The simulation experiment result indicates that when positive examples are 50% of total samples below,SCCIPM shows obvious advantages over other methods in solving the Single-Class Classification and has good classification performance.
出处
《计算机科学》
CSCD
北大核心
2011年第6期191-194,共4页
Computer Science
基金
国家自然科学基金(60675030
60875029)资助
关键词
集成金字塔模型
单分类
数据挖掘
Integrated pyramid model
Single-class classification
Data mining
