一种改进的C4.5算法  

An Improved C4.5 Algorithm

作　　者：李瑞[1] 程亚楠[1] 

机构地区：[1]大连交通大学软件学院,大连116028

出　　处：《科学技术与工程》2010年第27期6670-6674,共5页Science Technology and Engineering

基　　金：辽宁省自然科学基金(20072161)资助

摘　　要：为了提高C4.5算法的有效性,提出了一种改进的MB—C4.5算法。该算法主要改进了C4.5算法的分枝策略和属性选取的标准。把分类效果较差的分枝合并到分类效果较好的分枝中。引进一个平衡度系数,系数大小由决策者依靠先验知识或领域知识确定。MB—C4.5算法在提高重要属性的选择、减少无意义分枝、过度拟合等方面有一定提高。用该算法构造出的决策树进行分类更为准确、合理。对改进前后的算法用实例进行分析,说明MB—C4.5算法的有效性。To improve the effectiveness of C4.5 algorithm,an improved MB—C4.5 algorithm is introduced.The algorithm is mainly improved in the criterion of partitioning rules and attribution selection of the C4.5 algorithm：the branches which have poor appearances in classification are united into the ones which have good appearances in classification.A balanced coefficient is introduced and it can be fixed by decision maker according to priori intellectual and domain intellectual.MB—C4.5 enhances importance of test attribute selection,reduces the number of insignificant branches and avoid the appearance of over fitting.The classification is more veracious and rational by the decision tree made from the improved algorithm.And compared the improved algorithm to C4.5 algorithm by analyzing examples,to prove the efficiency of the improved algorithm.

关 键 词：C4.5算法 MB—C4.5算法 合并分枝 平衡度系数 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]