Enhanced Feature Selection Clustering Algorithm for Attribute Similarity in High Dimensional Data

Data collection is aggressive concept in data mining which is based on various attributes from dissimilar data sets. For some real world data, real time dataset portioning with abnormal behavioral class label instances is expensive and impossible to data presentation. Through user preferences, now a day’s data summarization based on clustering with different attributes is another aggressive concept. Traditionally clustering with multi-attribute framework was introduced to group multiple attributes to explore uncertain data for reliable data sets. In multi attribute similarity measure for uncertain data, feature selection is the factor to provide most matched and most useful features which produces compatible results from original set of features present in data sets. So feature selection algorithm is required to evaluate efficiency to form subset of features with respect to quality assurance for subset of features. In this paper, we proposed and implemented Enhanced Feature Selection based Clustering (EFSC) algorithm to evaluate above considerations. Our proposed method consists of two stages in implementation. In first stage, classify features into clusters using graph based theoretic approach. In second stage, identify most representative attribute which is most relate to selected attribute from each cluster to sub set of features. In this paper, we use Minimum spanning tree (MST) for effective clusters formation with respect to subset of features. EFSC is compare with some existing algorithms like FCBF, ReliefF, CFS, Consist, and FOCUS-SF with respect to chosen classifiers prior to and later than feature selection from subset of features. Our experimental results performed on company statistical data with text, image orientated data, and EFSC produces small subset of features with high accuracy and less time efficiency for real time data sets.