An Empirical Study on Evaluating Graph Based Clustering for HD Data Using Attribute Selection

V. Hemapriya; K.P.N.V. Satya Sree; K.V. Narasimha Reddy

doi:10.37591/joadms.v1i2.105

Abstract

An attribute subset selection can be showed as a process of identifying and eliminating or removing a number of irrelevant and surplus attributes (features) because irrelevant attributes do not give predictive accuracy and the surplus attributes provide the information that is already present in the other attributes. Attribute selection involves identifying a subset of the most useful attributes that produces the similar results as the final set of results. An attribute (feature) selection algorithm may be evaluated from two points of view. First one concerns the time required to get the subset of attributes and the second one concerns quality of the subset of attributes. Based on these criteria, graph-based clustering for attribute selection algorithm, GRACE is proposed. This algorithm works in two steps. In the first step, attributes are divided into clusters by using graph-theoretic clustering methods. In the second step, most similar attributes that are strongly related to the object classes are selected from each cluster from a subset of attributes. Attributes in different clusters are relatively independent. To ensure the efficiency of this algorithm, the authors implemented the minimum spanning tree clustering method.Keywords: Graph-based clustering, filter method, attribute subset selection