Finding banded patterns in data: The banded pattern mining algorithm

Abstract

The concept of Banded pattern mining is concerned with the identification of “bandings” within zero-one data. A zero-one data set is said to be fully banded if all the “ones” can be arranged along the leading diagonal. The discovery of a banded pattern is of interest in its own right, at least in a data analysis context, because it tells us something about the data. Banding has also been shown to enhances the efficiency of matrix manipulation algorithms. In this paper the exact N dimensional Banded Pattern Mining (BPM) algorithm is presented together with a full evaluation of its operation. To illustrate the utility of the banded pattern concept a case study using the Great Britain (GB) Cattle movement database is also presented.