非线性噪声数据集上基于随机森林的空缺值填补算法  

A MISSING VALUES FILLING ALGORITHM BASED ON RANDOM FOREST FOR NON-LINEAR NOISY DATASETS

作　　者：吴俊杰[1] 赵鹏[1] 

机构地区：[1]太原师范学院计算机系,山西太原030012

出　　处：《计算机应用与软件》2013年第7期51-53,共3页Computer Applications and Software

基　　金：国家自然科学基金项目(51174275);山西省高等学校科技项目(20121021);山西省社科联2012重点课题研究项目(SSKLZDKT2012055);山西省2011年科学技术发展计划项目(20110321024-02)

摘　　要：针对目前空缺值填补方法在非线性噪声数据集上填补效果不理想的问题,分标称属性和非标称属性两种情况提出一种基于随机森林的空缺值填补算法。该算法首先将空缺值看作决策属性,将其他属性值作为特征属性,然后利用随机森林算法进行空缺值的预测。由于随机森林算法具有良好的非线性数据拟合和抗噪声性能,因此该算法可以有效地提高空缺值的填补准确率。在UCI标准数据集和ORL人脸识别数据集上的对比实验充分说明了该算法较以往的填补方法更为有效。As present missing values filling algorithms are not satisfactory at filling for non-linear noisy datasets, the paper proposes a miss- ing values filling algorithm based on random forest for both nominal attributes and non-nominal attributes. The algorithm firstly regards the missing value as decision attribute and other attribution values as feature attributes; then by random forest algorithm it executes prediction for the missing value. Since the random forest＇algorithm is good at non-linear data fitting and anti-noise property, the proposed algorithm can effec- tively improve the filling rate of the missing value. Comparison experiments on UCI standard datasets and ORL face recognition datasets fully explains that the proposed algorithm is more effective than previous filling algorithms.

关 键 词：空缺值 非线性 噪声数据集 随机森林 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]