基于相关系数的决策树优化算法

通过分析ID3算法的基本原理及其多值偏向问题,提出了一种基于相关系数的决策树优化算法。首先通过引进相关系数对ID3算法进行改进,从而克服其多值偏向问题,然后运用数学中泰勒公式和麦克劳林公式的性质,对信息增益公式进行近似简化。通过具体数据的实例验证,说明优化后的ID3算法能够解决多值偏向问题。标准数据集UCI上的实验结果表明,在构建决策树的过程中,既提高了平均分类准确率,又降低了构建决策树的复杂度,从而还缩短了决策树的生成时间,当数据集中的样本数较大时,优化后的ID3算法的效率得到了明显的提高。

An optimized algorithm of decision tree based on correlation coefficients

Aiming at the problem of multi value bias in ID3 algorithm, we propose an optimized algorithm of decision tree based on correlation coefficients. Firstly, the correlation coefficients between the attributes are introduced to improve the ID3 algorithm, and in turn the multi value bias problem is overcome. Then the properties of Taylor formula and Maclaurin formula are adopted to simplify the information gain formula. The concrete data of examples prove that the optimized ID3 algorithm can overcome multi value bias problem. Experiments on the standard UCI data sets show that the optimized algorithm of decision tree not only improves the accuracy of average classification, but also reduces the complexity in building decision trees and thus reduces the generation time of decision trees. Besides, the efficiency of the optimized ID3 algorithm increases significantly for large scale samples.