Discovery of Intrinsic Clustering in Spatial Data

A fundamental task in knowledge discovery is the unraveling of clusters intrinsically formed in spatial databases. These clusters can be natural groups of variables, data-points or objects that are similar to each other in terms of a concept of similarity. They render a general and high-level scrutiny of the databases that can serve as an end in itself or a means to further data mining activities. Segmentation of spatial data into homogenous or interconnected groups, identification of regions with varying levels of information granularity, detection of spatial group structures of specific characteristics, and visualization of spatial phenomena under natural groupings are typical purpose of clustering with very little or no prior knowledge about the data. Often, clustering is employed as an initial exploration of the data that might form natural structures or relationships. It usually sets the stage for further data analysis or mining of structures and processes. Clustering has long been a main concern in statistical investigations and other data-heavy researches (Duda and Hart 1974; Jain and Dubes 1988; Everitt 1993). It is essentially an unsupervised learning, a terminology used in the field of pattern recognition and artificial intelligence, which aims at the discovery from data a class structure or classes that are unknown a priori. It has found its applications in fields such as pattern recognition, image processing, micro array data analysis, data storage, data transmission, machine learning, computer vision, remote sensing, geographical information science, and geographical research. Novel algorithms have also been developed arising from these applications. The advancement of data mining applications and the associated data sets have however posed new challenges to clustering, and it in turn intensifies the interest in clustering research. Catering for very large databases, particularly spatial databases, some new methods have also been developed over the years (Murray and Estivilli-Castro 1998; Miller and Han 2001; Li et al. 2006). To facilitate our discussion, a brief review of the clustering methods is first made in this section.