Dimension reduction using clustering algorithm and rough set theory

摘要: 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 valued dataset using the concept Rough Set Theory and clustering generate reduct. Here, projection based on two conditional Ci Cj is taken K-means Clustering applied it with K = distinct values decision attribute D obtain clusters. Also clustered into K-groups Indiscernibility relation D. Then connecting factor k combined (CiCj) respect calculated cluster sets set ACS {(CiCj$\rightarrow^{\hspace*{-2.5mm}^k} D$) all Ci,Cj ∈ C, Conditional set, (Decision attribute)} formed. Each element (CiCj$\rightarrow^{\hspace*{-2.5mm}^k} implies that together partition objects yields (k*100) % similar partitions as made Now an undirected weighted graph weights constructed ACS. Finally weight associated edges, attributes, called reduct generated. Experimental result shows efficiency proposed method.