基于非线性拟合方程的多变量决策树算法

根据数据属性间存在的线性相关和非线性相关影响决策树性能的特点,提出了一种用拟合回归建立决策树的算法,并利用这种相关性来提高分类能力.该算法选择了一个较优的属性子集,对此子集中的属性进行加权组合,用于构造决策树的节点,采用二次多项式来拟合两个属性间可能存在的相关性,从而构造出分类能力更强的决策树.研究中用UCI标准数据集对各种算法进行测试及比较,实验结果及分析表明此决策树算法具有良好性能.

A multi-variable decision tree algorithm based on nonlinear fitting equation

Linear and nonlinear correlations between features affect the ability of the decision tree. To exploit those correlations and improve classification ability, a new method was proposed to construct decision trees in which a quadratic fitted model was used. The algorithm selected approximatively optimal feature sets, and these features were weighted and assembled so as to construct a node of the decision tree. At the same time to model those possible linear and nonlinear correlations between features, features were fitted using quadratic fitted equations each time, then the datasets were partitioned by fitted results,and the subdataset is handled by recursive process to build a decision tree with better performance of classification. UCI standard data sets were tested in the research, and classification results of different algorithms show that the proposed decision tree algorithm has better performance than other tested decision tree algorithms.