Clustering the buildings in Istanbul by data mining approach

In this study, it is aimed to cluster the buildings according to floor, basement, geology, usage style and construction type attributes and to visualize those clusters by means of thematic mapping, based on existing data of buildings in Istanbul. First, related data was preprocessed before clustering in order to obtain more accurate results and reduce computing time. Preprocessing step consists of data integration, data cleaning, data reduction and normalization. It is the most time consuming step of data mining. Several kinds of software are used in each step, depending on their speed. These are MS Access, MS Excel, MapInfo, ArcMap and some little programs coded in C++. DBSCAN clustering algorithm was selected in this study due to some advantages over the other algorithms. It requires that user determines the parameters as in many other data mining algorithms. Seven clusters were obtained using the intuitively determined parameters within DBSCAN. Three of them were split into two parts in terms of mean value due to the high standard deviation of their floor attribute. Outliers (objects not a member of any cluster after clustering) were also used in creating a new cluster. Thus, totally eleven clusters are obtained. Distribution values of each attributes of each cluster are visualized in graphics for both whole objects and original objects. Finally, number of clusters within districts was visualized in graphics as well as their spatial distribution was visualized via thematic mapping.