Using Parametric Test to determine the Significance of Banded patterns in N-dimensional 0-1 dataset

Background: The identification of banded patterns in N-dimensional (ND) 0-1 dataset is one where all the elements in the dimensions are arranged such that the one’s entries are ordered about the center of dimensions. Reordering 0-1 datasets in order to determine banded patterns in data, allows for the identification of interesting pattern that are hidden in the data. The challenge is whether or not the identify banded patterns are significant. Previous work on the significance of banded patterns using parametric test was aimed at 2D and 3D banding algorithms, using a single banding algorithm in each case, which meant no work on the significance of banded patterns using different banding algorithms with different dimensions. Aim: To this end, this paper presents a comparison between ND banding algorithms for 2D and 3D. Method: The approach is to use parametric test on synthetic data, UCI and the real datasets taken from the cattle tracing system (CTS). The ND banding algorithms considered for 2D are:2D banding, barycentric (BC) and 2D sort, and for 3D are: exact-Euclidean, exact-Manhattan variations and the approximate. Results: The experimental results presented shows the significance of banded patterns with p value less than 0.05. However, the post-hoc test result shows a statistically significant difference between 2D banding and BC, 2D banding and 2D sort, exact-Euclidean and exact-Manhattan, exact-Euclidean and the approximate but no significant difference between BC and 2D sort as well as exact-Manhattan and the approximate.