DBSCAN and CLARA Clustering Algorithms and their usage for the Soil Data Clustering

Abstract

In this paper two clustering algorithms DBSCAN and CLARA were applied over the pedological database of Montenegro. Both algorithms clusterize data based on their density distribution. DBSCAN enables discovering clusters of arbitary shapes, without domain knowledge. On the other hand, CLARA forms clusters of approximatly equal size and shape for databases with uniformly spaced data. The used databases is composed of chemical and mechanical-physical parameters of soil samples. There are no clear transitions between different types of soil and large differences in values of their parameters at the boundary points of the clusters. Thus, CLARA is proved to be better for clustering pedologic data, which is confirmed by means of simulations. The results obtained by the CLARA are comparable with the results obtained by the analysis of soil in Montenegro by the expert.