Discovering Spatial Regions of High Correlation

• Published in: 2016 IEEE 16th International Conference on Data Mining Workshops (ICDMW) pp. 1082 - 1089
• Main Authors: Agarwal, Prerna, Verma, Richa, Gunturi, Venkata M. V.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01-12-2016
• Subjects: Algorithm design and analysis; Area measurement; Clustering algorithms; Co-location Pattern Mining; Correlation; Data mining; Density measurement; Regional Co-location Mining; Spatial Data Mining; Transportation
• ISSN: 2375-9259
• DOI: 10.1109/ICDMW.2016.0156

Given a set of events of two different types (e.g. locations of crime incidents/road accidents) in geographic space and minimum density and area thresholds, spatial regions of high correlation discovery (RHC) aims to determine rectangular-shaped areas of high correlation between two event types. RHC discovery is important to many fields like transportation engineering, criminology, and epidemiology. Designing a scalable algorithm for RHC discovery is challenging mainly because of non-monotonicity of popular spatial statistical interest measures of association between events, one such measure being the cross-K function. This challenge makes Apriori-based pruning algorithms inapplicable. The large enumeration space is another challenge. To address these limitations, we propose a cross-K inspired interest measure and using that, a novel algorithm for RHC discovery. Real crime data is used for a case study to present the output of our algorithm. Experimental evaluation is done to show that the proposed algorithm cuts down on computation substantially as compared to the naive approach.