A Systematic Mapping Study on Ensemble-Based Classifier

Abstract

The performance of classifiers is essential for the outcome of machine learning algorithms. A practical approach to significantly increase the performance of a classifier is through an ensemble-based classifier. Therefore, ensemble of classifiers must be examined. The purpose of this study is to provide an overview of recent machine learning research (2016–2020) in ensemble-based classifiers, with the primary goal of understanding what research has been carried out among the machine learning community that implements ensemble-based classifiers in their studies. We performed a systematic mapping study to investigate ensemble-based classifiers in machine learning research. A total of 149 prominent papers published from 2016 to 2020 implement this method using various dataset types, and each of these papers addresses the issues encountered with the ensembled classifier. These studies were classified according to the ensemble-based classifier techniques, issues and types of datasets. As a result, the most favorable ensemble technique used by many researchers is the hard-level approach. However, there is a lack of research that addresses the issue of multi-class imbalanced data. Thus, indicates a need for more studies in handling multi-class imbal-anced data. We also propose a decision making flowchart which may aid researcher in making informed decisions on how best to analyze data. The results suggest that researchers should give more attention to the combining techniques of ensemble-based classifiers in order to achieve an ample overall performance, and clearly there is a need for more studies on the issue of multi-class imbalanced data.