An Empirical Comparison of Ensemble and Hybrid Classification

The application of Data mining has proved to be successful in almost all the fields including medical domain. Medical data mining is the process of extracting useful knowledge and hidden patterns from medical data. This paper proposes a hybrid model for classifying Cleveland Heart dataset with hybrid feature selection and compares the performance with the base classifiers and ensemble classifiers. The model is developed in four stages. In the initial stage, Cleveland Heart dataset selected from the UCI repository is cleaned by deleting all the instances with missing values. In the second stage Fuzzy and Rough Set is used in a cascaded fashion for relevant feature extraction. In the third stage the resultant dataset was clustered into two segments using K-means and incorrectly clustered samples were eliminated to get final samples. Finally, the correctly clustered samples from the previous stage was trained with 5 different classifiers to build the final classifier model using 10 fold cross validation. Experimental results proved that proposed hybrid model showed enhanced classification accuracy compared to base classifiers and ensemble classifiers. It yielded highest accuracy of 99.54%.