A Neighbourhood Rough Set Based Clustering Algorithm and its Applications

Abstract

B. K. Tripathy , Akarsh Goyal

The process of Data clustering puts similar objects into the same group. The attributes under consideration may be numerical or categorical. A type of partition attribute based clustering algorithm dealing with categorical attributes only through rough sets was started in 2007 with the Min-Min Roughness (MMR) algorithm due to Parmar et al. It was generalised to the Min Mean roughness (MMeR) algorithm dealing with heterogeneous attributes by Kumar et al in 2009. Here, the numeric attributes are transformed into categorical ones. It was further improved through the Min Standard Deviation Roughness (SDR) algorithm and the Standard deviation Standard Deviation Roughness (SSDR) algorithm by Tripathy et al in 2011. A natural approach to deal with both types of attributes together, which extends all these algorithms is the Min Mean Neighbourhood Roughness (MMeNR) algorithm, which can be applied to uncertainty based heterogeneous attributes. This algorithm is shown to be superior to the above algorithms through the computation of F-measure and bench marked data sets like Teacher Assistant Evaluation Data Set and Acute Inflammations Data Set from UCI repository. Spatial datasets like the Forest Fire and the Abandoned Mine Land Inventory Data are used to show the superiority of MMeNR among all.

PDF

Share this article