ENSEMBLE SELECTION AND COMBINATION BASED ON COST FUNCTION FOR UCI DATASETS

It is well-known that the classification performance of any single classifier is outperformed by a multiple classifier approach or an ensemble process that incorporates results from different base classifiers. However, even though they have the potential to achieve greater classification precision, their vast number of base classifiers has greatly influenced ensemble methods. In the ensemble process, the selection and combination of appropriate and varied classifiers is a daunting task. In the previous work, we, therefore, suggested a new soft ensemble selection and combination approach (SSSC) to identify the best subset of heterogeneous ensemble team of classifiers and demonstrated the potential of our proposed algorithm to minimise a large number of classifiers while at the same time generating the highest predictive precision for consumer churn data sets. This paper extended the earlier work with the goal of evaluating whether the proposed SSSC model works well with the other UCI repository benchmark data sets. The findings of the experiments demonstrated that the developed model resulted in the improvement of the chosen UCI data sets' prediction accuracy. Based on the results of the experiments, it indicates that the prediction accuracy of the proposed SSSC outperformed other single classifiers and ensemble methods for Liver Disorder, Hepatitis and Breast Cancer data sets. This work has shown that the proposed SSSC is able to search for a minimal number of classifiers in the repository of the ensemble while at the same time enhancing the precision of the classification of the chosen UCI data sets