Correlation Coefficient Based Candidate Feature Selection Framework Using Graph Construction

Authors : Sai Prasad POTHARAJU, Marriboyina SREEDEVI

Pages : 775-787

View : 22 | Download : 7

Publication Date : 2018-09-01

Article Type : Research Paper

Abstract :Selection of strong features is crucial problem in machine learning. It is also considered as an inescapable exercise to minimize the number of variables available in the primary feature space for finer classification performance, decrease computation complexity , and minimized memory utilization. In this current work, a novel structure using Symmetrical Uncertainty insert ignore into journalissuearticles values(SU); and Correlation Coefficient insert ignore into journalissuearticles values(CCE); by constructing the graph to select the candidate feature set is presented. The nominated features are assembled into limited number of clusters by evaluating their CCE and considering the highest SU score feature. In every cluster, a feature with highest SU score is selected while remaining features in the same cluster are disregarded. The presented methodology was investigated with Teninsert ignore into journalissuearticles values(10); well known data sets. Exploratory results assures that the presented method is out pass than most of the traditional feature selection methods in accuracy. This framework is assessed using Lazy, Tree Based, Naive Bayes, and Rule Based learners.
Keywords : Feature Selection, Correlation Coefficient, Classification, Machine Learning, Symmetrical Uncertainty