Domain-driven density based clustering algorithm

Abstract

Due to increasing demand to derive knowledge from data, there is need for efficient data mining algorithms. We have proposed an algorithm to get clusters of arbitrary shapes with the help of Density Based clustering algorithms. Density Based clustering algorithms need Epsilon Value (Eps) and Minimum Points Value (MinPt) to create clusters. Hence in this paper, a method is proposed which accepts the domain knowledge about the dataset as an input and calculation of Eps and MinPt is automated which helps to make the data certain to some extent. Domain knowledge adds some relevance to data, hence this data with knowledge will never be ignored during clustering. In this method, we first create grids for dataset as per user’s requirement then it derives the default Eps and MinPt which are inputs for Density Based Clustering Algorithm for Large Datasets (DBSCALE) algorithm. The results taken after implementation shows the proposed method gives better clusters.