Intelligent feature selection using hybrid based feature selection method

Feature selection and reduction has assumed the position of a leading approach for many preprocessing step in machine learning. It is widely used as preprocessing in classification due to an exponential growth in data set as well as in feature set. The aim of this paper is to contribute in the domain of feature selection, which directly reduces the complexity and speed up the learning algorithm by improving the predictive accuracy. In machine learning as the dimensionality increases, the classifiers performance increases until the optimal number of features is reached. Further increasing the dimensionality without increasing the number of training samples results in a decrease in performance, and increase in the complexity and computational time of the classifier. This degrades the overall performance of the classifiers. In this work, a new model is proposed, hybrid based feature selection (HBFS) that identifies relevant features with respect to labels as well as eliminate the redundancy between relevant features. Features selection under the proposed method is compared with the widely used techniques such as correlation based feature selection, mutual information based feature selection, Kullback Leibler based feature selection and improves the accuracy by 5.75%. The proposed model is evaluated using 10 fold cross validation technique. The results show that the proposed method not only reduces the number of features but also achieves accuracy up to 87.01%.