CMStatistics 2020: Start Registration
View Submission - CMStatistics
Title: A cross-entropy method for spatial clustering of binary data Authors:  Nishanthi Raveendran - Macquarie University (Australia) [presenting]
Georgy Sofronov - Macquarie University (Australia)
David Bulger - Macquarie University (Australia)
Abstract: Spatial data is very often heterogeneous, indicating that there may not be a unique statistical model describing the data. To overcome this issue, the data can be segmented into several homogeneous regions (or domains). Identifying such homogeneous domains and their boundaries is called spatial clustering (or segmentation) in spatial statistics. Spatial clustering is commonly used in many different fields, including epidemiology, criminology, and ecology. The focus is on spatially correlated binary data indicating the presence or absence of plant species observed over a two-dimensional lattice. Factoring out cluster label permutations yields a non-Euclidean model space. To address this, a hybrid of the Cross-Entropy method with a genetic algorithm is proposed. Voronoi tessellation is used to estimate the cluster centres and boundaries of such domains. The results illustrate that the proposed algorithm is effective in identifying homogeneous clusters in spatial binary data.