Dimension Reduction Using Clustering Algorithm and Rough Set Theory

Abstract

In real world, datasets have large number of attributes but few are important to describe them properly. The paper proposes a novel dimension reduction algorithm for real valued dataset using the concept of Rough Set Theory and clustering algorithm to generate the reduct. Here, projection of dataset based on two conditional attributes C _i and C _j is taken and K-means Clustering algorithm is applied on it with K = number of distinct values of decision attribute D of the dataset to obtain K clusters. Also the dataset is clustered into K-groups using Indiscernibility relation applied on the decision attribute D. Then the connecting factor k of combined conditional attributes (C _i C _j) with respect to D is calculated using two cluster sets and attribute connecting set ACS = {(C _i C _j $\rightarrow^{\hspace*{-2.5mm}^k} D$) for all C _i,C _j ∈ C, Conditional attribute set, and D (Decision attribute)} is formed. Each element (C _i C _j $\rightarrow^{\hspace*{-2.5mm}^k} D$) ∈ ACS implies that C _i and C _j connecting together partition the objects that yields (k*100) % similar partitions as made on D. Now an undirected weighted graph with weights as the connecting factor k is constructed using attribute connecting set ACS. Finally based on the weight associated with edges, the important attributes, called reduct are generated. Experimental result shows the efficiency of the proposed method.