一种基于双聚类的缺失数据填补方法  

Novel approach for missing data imitation based on biclustering

作　　者：郝胜轩 宋宏[1] 周晓锋[1] 

机构地区：[1]中国科学院沈阳自动化研究所,沈阳110016 [2]中国科学院大学,北京100049

出　　处：《计算机应用研究》2015年第3期674-678,共5页Application Research of Computers

基　　金：国家重大科技专项基金资助项目(2011ZX02507-006)

摘　　要：针对现实数据集的数据缺失问题,提出了一种基于双聚类的缺失数据填补新方法。该算法利用双聚类簇内平均平方残值越小簇内数据相似性越高的这一特性,将缺失数据的填补问题转换为求解特定双聚类簇最小平均平方残值的问题,进而实现了数据集中缺失元素的预测;再利用二次函数求解极小值的思想对包含有缺失数据的特定双聚类簇最小平均平方残值的问题进行求解,并进行了数学上的分析证明。最后进行仿真验证,通过观察UCI数据集的实验结果可知,提出的算法具有较高的填补准确性。In view of the problem of the lack of realistic data sets, this paper proposed a novel imputation method based on bi- clustering is proposed to solve the missing data problem. Firstly, the proposed method transformed the problem of imputing missing data into the problem of specific bicluster＇ minimum mean squared residue, which utilized the characteristics of the bicluster data that the smaller bicluster＇ s mean squared residue the higher similarity, thus the proposed method could predict the missing data in data sets. Secondly, it employed a solving minimization strategy of quadratic function to solve the problem of specific bicluster＇ s minimum mean squared residue, and gave the corresponding mathematical proof. Finally, it executed simulation and verification, and gave the resuhs of UCI data sets show that the proposed imputation method has higher accuracy compared with other imputation methods.

关 键 词：缺失数据填补 双聚类 双聚类数据填补 数据清洗 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]