A cluster validity for spatial clustering based on davies bouldin index and Polygon Dissimilarity function

• Published in: ICIC : 2017 second International Conference on Informatics and Computing : 1-3 November 2017 pp. 1 - 6
• Main Authors: Karo, Ichwanul Muslim Karo, MaulanaAdhinugraha, Kiki, Huda, Arief Fatchul
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.11.2017
• Subjects: Clustering algorithms; Clustering methods; Davies-Bouldin Index; Euclidean distance; Indexes; Partitioning algorithms; PDF; Silicon; Spatial Clustering; Spatial databases
• DOI: 10.1109/IAC.2017.8280572

Spatial clustering is most powerfully technology to spatial data mining. One of impartant part on spatial clustering is cluster validity and closely related with spatial dissimilarity. Dissimilarity function limitation makes cluster validity of spatial clustering become one of the most important issues on cluster analysis. However, traditionally cluster validity is fail and not fair to measure inter and intra cluster of region dataset. Main subject of this paper is a cluster validity for spatial region clustering by using modified of Davies Bouldin index with Polygon Dissimilarity function (PDF), called DB P . The DB P comprehensively combines both the spatial and the non-spatial attributes that exist within the datasets. To evaluate DB P , It was compared with other cluster validity (e.g Silhouette Index and Gap Static). The DB P can measure intra and inter cluster by using spatial dissimilarity function. In addition, we specifically investigate the effectiveness of our cluster validity in a spatial clustering application using a partitional clustering technique (e.g. CLARANS) using dummy region dataset. DB P has highest compactness than gap and silhouette index for best cluster. Moreover, DB P makes sense than silhouette index and Gap Static for spatially joint cluster.