CURRENT ISSUES IN ENSEMBLE METHODS AND ITS APPLICATIONS

• Published in: Journal of Theoretical and Applied Information Technology Vol. 81; no. 2; p. 266
• Main Authors: Baba, Nura Muhammad, Makhtar, Mokhairi, Fadzli, Syed Abdullah, Awang, Mohd Khalid
• Format: Journal Article
• Language: English
• Published: Islamabad Journal of Theoretical and Applied Information 01.11.2015
• Subjects: Accuracy; Classifiers; Construction; Information technology; Mathematical models; Optimization; Strategy; Tasks
• ISSN: 1817-3195

This paper reviewed the current state-of-the-art of optimization of ensemble methods so as to provide us with a better direction of how we will conduct our research in the future. The primary aim of ensemble method is to integrate a set of models that are used for solving different tasks so as to come up with enhanced composite global model, which produces higher accuracy and reliable estimate than what can be achieved through a single model. Diversity, combination strategies, number of based classifiers, types of ensemble, and performance measures are the key factors to be considered in the build of committees. When the numbers of base classifiers become huge, ensemble methods incurred high storage space and computational time, selective ensemble is proposed by most literatures to solve these problems. In terms of optimization techniques, multi-objectives techniques have become the better ones to use due to their efficiency in terms of optimization process and they provide a set of near optimal solution instead of just a single solution. When comparing the performance of ensemble methods, most of the time, accuracy alone cannot differentiate which classifiers perform best; for this reason, other performance measures such as AUC, F-measure, TPR, TNR, FPR, FNR, RMSE were used. Based on the reviewed literatures, we concluded that in our proposed methodology we would come up with a new method for comparing and searching for relevant classifiers from a collection of models that would be used as a model for predicting the quality of water to achieve higher performance rate than other previous work.