ITGC: Information-theoretic Grid-based Clustering

Grid-based clustering algorithms are well-known due to their efficiency in terms of the fast processing time. On the other hand, when dealing with arbitrary shaped data sets, density-based methods are most of the time the best options. Accordingly, a combination of grid and density-based methods, where the advantages of both approaches are achievable, sounds interesting. However, most of the algorithms in these categories require a set of parameters to be specified while usually it is not trivial to appropriately set them. Thus, we propose an Information-Theoretic Grid-based Clustering (ITGC) algorithm by regarding the clustering as a data compression problem. That is, we merge the neighbour grid cells (clusters) when it pays off in terms of the compression cost. Our extensive synthetic and real-world experiments show the advantages of ITGC compared to the well-known clustering algorithms.