Quality Assessment in Spatial Clustering of Data Mining

Because of the use of computers and its advances in scientific data handling and advancement of various geo and space borne sensors, we are now faced with a large amount of data. Therefore, the development of new techniques and tools that support the transforming the data into useful knowledge has been the focus of the relatively new and interdisciplinary research area named “knowledge discovery in spatial databases or spatial data mining”. Spatial data mining is a demanding field since huge amounts of spatial data have been collected in various applications such as real-estate marketing, traffic accident analysis, environmental assessment, disaster management and crime analysis. Thus, new and efficient methods are needed to discover knowledge from large databases such as crime databases. Because of the lack of primary knowledge about the data, clustering is one of the most valuable methods in spatial data mining. As there exist a number of methods for clustering, a comparative study to select the best one according to their usage has been done in this research. In this paper we use Self Organization Map (SOM) artificial neural network and K-means methods to evaluate the patterns and clusters resulted from each one. Furthermore, the lack of pattern quality assessment in spatial clustering can lead to meaningless or unknown information. Using compactness and separation criteria, validity of SOM and K-means methods has been examined. Data used in this paper has been divided in two sections. First part contains simulated data contain 2D x,y coordinate and second part of data is real data corresponding to crime investigation. The result of this paper can be used to classify study area, based on property crimes. In this work our study area classified into several classes representing high to low crime locations. Thus, accuracy of region partitioning directly depends on clustering quality.