基于异质矩阵完全的缺失数据恢复混合集成算法

Mixed Ensemble Heterogeneous Matrix Completion for Missing Value Estimation

缺失数据广泛存在于现实世界中,它对后续的数据分析有很大的影响,有可能导致结果完全错误。近年来,很多基于压缩传感理论的矩阵完全算法被提出并用于缺失数据恢复,但不同的算法在不同的数据集上产生的结果有很大不同,都有自己的优缺点和适用场景。为此提出一种基于异质矩阵完全算法和最大多样性的集成策略的混合集成学习算法,实验结果表明,此算法在不同的数据集上优于那些单个算法。

The problem of incomplete data is ubiquitous in real world and has a significant effect on the application of data analysis method and final conclusion. Recently, many matrix completion algo- rithms based on compress sensing are proposed, while they all have respective advantages and disad- vantages, and every approach vary drastically on different datasets and their preferences and potential limitations are special. In this paper, a mixed ensemble method is proposed based on heterogeneous matrix completion algorithms and ensemble strategy with high--level diversity. The experiment re- sult shows that ensemble method outperforms all single matrix completion algorithms in different datasets.