一种自整定权值的多分类器融合方法  

A Method of Multiple Classifier Fusion with Self-Adjusting Weights

作　　者：米爱中[1,2] 郝红卫[2] 郑雪峰[2] 涂序彦[2] 

机构地区：[1]河南理工大学计算机科学与技术学院,河南焦作454000 [2]北京科技大学信息工程学院,北京100083

出　　处：《电子学报》2009年第11期2604-2608,F0003,共6页Acta Electronica Sinica

基　　金：国家自然科学基金(No.60675006);"十一五"国家科技支撑计划(No.2006BAI03A09);河南省教育厅自然科学研究计划(No.2008A520010)

摘　　要：本文提出一种自整定权值的融合方法.该方法使用混淆矩阵来衡量分类器性能,并根据分类器输出情况自适应地为各分类器赋予权值,可靠的决策结果获得较大的权值,从而提高决策模板的可信度.对易于被错误分类的样本,在利用其与决策模板的相似性信息的同时,结合它周围的训练样本信息做出判断.通过与DT方法在KDD’99入侵检测数据集和UCI数据库中的8个数据集上的实验对比,表明本文方法具有更好的分类性能.A fusion method with self-adjusting weights is proposed, which measures the classifier performance by the confusion matrix,and self-adaptively assigns weights to classifiers based on their outputs. Bigger weights are assigned to reliable outputs so that the decision templates are more credible. For a sample which is prone to be misclassified,besides the similarity between it and the decision templates, the information of the training samples around it are included to make a decision. Experiments were done on the KDD＇ 99 intrusion detection dataset and 8 datasets from the database UCI to compare the proposed method with the DT method. The experimental results show the presented method has a better classification oerformance.

关 键 词：多分类器系统 决策融合 混淆矩阵 决策模板 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]