摘要
单变量决策树算法生成的决策树具有规模庞大、规则复杂且不易理解的不足。采用粗糙集相对核、加权粗糙度的概念和类别因子相结合的方法,提出一种新的决策树生成算法。对于即将生长的节点,若节点样本的类别因子大于给定阈值,则停止生长该节点,如此就有效地避免了划分过细的问题。通过实验说明,该算法比传统的ID3算法生成的决策树更简单、更易于理解、抗噪声能力更强。
The decision tree generated by univariate decision tree algorithm has defects of huge in size,complicated rules and difficult in comprehensibility.The algorithm proposed in the paper generates decision tree by integrating relative core and weighted roughness with category factor.The node to be grown will stop growing if whose category factor is bigger than the given threshold,so it avoids the problem of dividing too fine.Experiment indicates that the decision tree generated by this algorithm is simpler,more understandable and more antinoise than the one generated by ID3.
出处
《计算机应用与软件》
CSCD
2010年第6期95-97,共3页
Computer Applications and Software
基金
辽宁省教育厅基金项目(20031066)
关键词
单变量决策树
多变量决策树
加权粗糙度
类别因子
相对核
Univariate decision tree Multivariate decision tree Weighted roughness Category factor Relative core
