Comparative study of attribute reduction on arrhythmia classification dataset

The research presented in this paper is focused on comparative study of various attribute selections as one of preprocessing methods used in world machine learning applications. Using UCI arrhythmia dataset, nine combination of attribute selection, based on search methods (Best First, Genetic Search and PSO Search) and attribute evaluator (CfsSubsetEval, ConsistencySubsetEval, and RSARSubsetEval) are tested and compared. Those data of attribute reduction results are then classified by using eight classifiers (Naive Bayes, Bayes Net, MLP Classifier, RBF Classifier, Jrip, PART, J48 and Random Forest). The best overall results are achieved by the combination of Best First and CsfSubsetEval which has the accuracy of 81% when it is tested with RBF Classifier. PSO Search methods was also found not very effective to generate high quality subsets.