基于改进[FK(W][BG(][BHDWG1,WK*2/3W][ZB(][BHDWG*3,WKW] χ [ZB)W][BG)W][FK)] 2统计的数据离散化算法

在基于χ2统计独立性的离散化算法中,自由度与期望频数的选取直接影响χ2计算的准确性,从而影响离散化的性能.为此,提出了一种基于改进χ2统计的数据离散化算法,提高了基于统计独立性离散化算法的质量.首先,分析了χ2函数中自由度选取的不足,给出了自由度选取的修正方案;其次,根据数据类分布等特点,提出了期望频数的改进方案,克服了不同数据集赋予相同期望频数的缺陷,提高了χ2计算的准确性.实验结果表明,改进的方法显著提高了C4.5决策树与Naive贝叶斯分类器的学习精度.

A data discretization algorithm based on improved chi-square statistic

The selection of degree of freedom and expected frequency directly affects the accuracy of chi-square calculation in discretization algorithms based on chi-square statistical independence.This will affect the performance of discretization.A data discretization method based on improved chi-square statistic is proposed.It improves the quality of discretization algorithm based on statistical independence.Firstly,the deficiency of the selection of degree of freedom in chi-square function is analyzed,and a modified scheme for selection of degree of freedom is given.Secondly,an improved scheme for expected frequency is proposed according to data class distribution,which overcomes the deficiency that different datasets have the same expected frequency.This improves the accuracy of chi-square calculation.The experimental results show that the improved algorithm improves the learning accuracy of C4.5 decision tree and Naive Bayes classifier.