摘要
引入图的误分类代价矩阵,选取以最小误分类代价为目标的加权子图作为图样本的特征属性,建立起图的决策树桩分类器,进行集成学习,得到一个对新图进行分类的判别函数.在生成候选子图时,利用子图的超图增益值具有上界的性质来裁剪增益值比较小的候选子图,从而减少候选子图数量,提高算法效率.实验结果表明,所提算法比其他图分类算法的误分类代价更小.
Introduce cost - matrix of graph misclassification, select the weighted sub - graph based on the least misclassification cost as the attribute of graph, then build up decision stump classifier and en- semble learning, finally obtain a classify critical function to classify a new graph. And use the upper bound of super graph for reducing the number of candidate sub - graphs. Experimental results show that our algorithm performs better than another graph classification algorithm in reducing misclassifica- tion cost.
出处
《福州大学学报(自然科学版)》
CAS
CSCD
北大核心
2012年第3期316-321,共6页
Journal of Fuzhou University(Natural Science Edition)
基金
国家自然科学基金资助项目(61070020)
福建省高等学校新世纪优秀人才支持计划资助项目(XSJRC2007-11)
关键词
代价敏感
图分类
集成学习
图挖掘
频繁模式
cost sensitivity
graph classification
ensemble learning
graph mining
frequent pattern