Data Matrix Completion Based on Pattern Classification

Abstract

In recent years, with the rapid development of big data technology, the matrix completion is often used for data recovery, and how to improve the accuracy of matrix completion is a key issue. This paper proposes a matrix completion method based on pattern classification, called PCRE, to improve data recovery performance. Since the hidden similarity within the data is a significant factor affecting the overall performance, the method PCRE uses non-negative matrix decomposition to extract the patterns of the data and accordingly rearranges the data matrix to fit for the matrix completion. Experiments are conducted by using PM 10 monitoring data collected by 34 sensors in Beijing in 2019 (totally 351 days). The results show that, compared with existing methods, PCRE improves the accuracy of data recovery with a shorter computation time.