RegK-Means: A Clustering Algorithm Using Spatial Contiguity Constraints for Regionalization Problems

Regionalization embraces many problems in which the recognition and analysis of patterns in geographic regions is necessary. Clustering techniques and some optimization methods offer algorithms and tools to facilitate this task, and AZP (Automatic Zoning Problem) is one of these optimization methods. In this work, we use semi-supervised clustering by adapting K-Means for regionalization, which was not previously explored. For this purpose, we specified contiguity constraints based on a neighborhood representation. The proposed algorithm is called RegK-Means. Experimental results using three real datasets show improvement in intra-cluster variance, minimization of objective function and computational time improvement compared to AZP.