一种新型的模型树算法研究及应用

On A New Model Tree Algorithm and Its Application

决策树算法训练速度快、结果易于解释,但在实际应用中其分类精度难以满足业务要求。为了提高决策树算法的精度,基于LogitBoost算法的优点,对决策树C4.5算法进行了改进。在决策树的叶节点上应用LogitBoost算法建立叠加回归模型,得到一种新型的模型树算法-LCTree算法。通过11组UCI数据集试验,经分析比较,证明LCTree算法比其他算法更有效。将该算法应用于电信客户离网预警系统建模,结果表明,该算法可有效地分析客户特征,精确地预测离网客户。

Decision tree algorithm has high training speed and easy understandable result, but it has the disadvantages of limited precision resulting and is unable to satisfy the need of application in businesses. Basal on the advantages of LogitBoost algorithm, a new model tree algorithm called LCTree algorithms is proposed. LCTree algorithm is an improved algorithm building Logistic model on the leaves of C4. 5 tree. The experiments of the LCTree algorithms show better precision than others algorithms. The application in classification modeling of telecommunications data set got more effective results than C4. 5 does.