Data matrix compression by using co-clustering

A two dimensional data matrix has been widely used in many applications. The lossless compression of data matrix not only brings benefits for storage but also for network transmission. In this paper, we propose a novel data-mining-based compression approach consisting of three steps: reordering and grouping data matrix columns and rows by co-clustering; post-processing to further expose redundancy in data matrix; data compression by a standard compressor. The inverse transform of co-clustering is very fast and simple, which facilitates matrix uncompression. We tested the approach on a synthetic dataset and five UCI real-life datasets. The experimental results suggest that our approach can improve compression rates at least 24% and up to 68%. The results also show that the time cost of the approach is linearly proportional to data matrix size, which is faster than other competition methods.