Fast Identification of Interesting Spatial Regions with Applications in Human Development Research

Large-scale demographic datasets with spatial information provide a rich platform for human development research. Much emphasis is often placed on understanding deviations from dataset-level behavior across demographic attributes within spatially coherent regions, since those could point to a local condition worth addressing through regional policies, or at the other extreme, a less known success story that offers new learnings. Inspired by such scenarios, we build upon domain knowledge from HDR to devise an interestingness scoring for spatial regions and formulate the computational task of interesting spatial region identification. Accordingly, we develop a taxonomic organization of spatial regions and formulate bounds on interestingness scores, which are then leveraged to develop an efficient technique to address the task. Our search method is empirically evaluated over two real-world datasets, and is seen to record orders of magnitude of response time improvements over region enumeration. The absolute response times and the memory overheads of our approach are seen to be within highly desirable ranges, establishing the effectiveness of our solution for the task.