Year: 2025
Type: conference-paper
Abstract: Spatial colocation pattern mining identifies spatial features frequently found near each other and plays a vital role in applications such as epidemiology and crime analysis. While prior research has focused on computational challenges like handling exponential candidate growth and costly spatial relationship checks, fewer studies address the varying interestingness of colocation patterns across regions. What is considered an interesting pattern in one country may not be relevant in another due to contextual and cultural differences. To address this, we propose a novel approach that analyzes the data in each region to determine the best estimate of optimal spatial neighborhood relationship constraints. We evaluate our method using the Global Terrorism Dataset and demonstrate through case studies that proximity alone is insufficient for defining consistent spatial relationships across regions. Additionally, we compare KDTree and BallTree for neighborhood queries and find that KDTree provides a faster runtime by an average of 11.29 seconds, making it a more efficient choice for large-scale spatial colocation mining.
Source: ACM Southeast Conference (ACMSE)
Authors: Abigail Kelly, Arpan Man Sainju, Dipesh Shrestha, Ramchandra Rimal
Institution: Middle Tennessee State University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Funder: National Science Foundation
OpenAlex ID: W4410206499