A multi-relational approach to spatial classification

摘要: Spatial classification is the task of learning models to predict class labels based on features entities as well spatial relationships other and their features. data can be represented multi-relational data, however it presents novel challenges not present in problems. One such problem that are embedded space, unknown a priori, part algorithm's determine which important what properties consider. In order when two spatially related an adaptive non-parametric way, we propose Voronoi-based neighbourhood definition upon literals built. Properties these neighbourhoods also need described used for purposes. Non-spatial aggregation already exist within framework, but sufficient comprehensive classification. A formal set additions mining framework proposed, able represent aggregations literals. These allow capturing more complex interactions occurrences trends. efficiently perform rule exploit powerful multi-processor machines, scalable parallelized method capable reducing runtime by several factors presented. The compared against existing methods experimental evaluation real world crime dataset demonstrate importance advantages parallelization.