Classification is a supervised machine learning procedure in which the effective model is constructed for prediction. The accuracy of classification mainly depends on the type of features and the characteristics of the dataset. Feature selection is an efficient approach in searching the most descriptive features which would contribute to the increase in the performance of the inductive algorithm by reducing dimensionality and processing time. In the present work a hybrid embedded feature selection algorithm with class label refining and handled numeric class problem in classifier are implemented. A novel feature selection algorithm based on ranker search optimization method and ensemble genetic search for selecting the appropriate features and class label refining for correcting misclassified instances from the dataset have been done. By modelling this approach, it reaches a near global optimal solution set of features. Hybrid feature selection with enhanced C4. 5 and naïve bayes classification can handle numeric class to achieve better performance. The efficiency of this method is demonstrated by comparing with the other existing methods in terms of accuracy, number of features selected and ability to handle numerical class values. Experimental results on datasets reveals that the proposed algorithm increases the classifier accuracy with less error rate and the quality of results are comparable.

Data Mining Hybrid Feature Selection Classification Decision Tree Accuracy