Download This PaperDiversity Matrix Based Performance Improvement for Ensemble Learning Approach
Hybrid Computational Intelligence : Research and Applications, ISBN 9781138320, 253 - CAT#K391719, CRC Press, Taylor and Francis Group.
33 Pages Posted: 26 May 2020
Date Written: October 1, 2019
Abstract
The ensemble learning approach is one of the widely used machine learning technique in classification problems. There are different types of ensemble classifiers and its combining techniques in use. In this chapter, a bagging ensemble technique is used on a motor-imagery EEG signal for brain-states discrimination problem. The majority-voting technique is implemented at the end for combining different predictors and to compute the final decision. We have observed through a rigorous empirical means that diversity in decision boundaries obtained from different learners plays an important role in the performance of the ensemble. In that direction, we have proposed a new diversity matrix based pruning technique which selects a subset of predictors from the actual set of predictors obtained from the ensemble.The aim of this chapter is that the selected subset of predictors (basically the predicted decision 2 - Diversity Matrix based Performance Improvement for Ensemble Learning Approach class vectors) performs better than the actual set. Using different configuration, we have empirically shown in the chapter that our proposed method performs quite satisfactory in its maiden variant. The best performing mean- improvement and best improvement in terms of ac-curacies are 01.75% (mean of 5 × 100 runs) & 07.14% respectively.
Keywords: Ensemble learning, Bagging, Diversity matrix, Majority voting, Classification, Accuracy
JEL Classification: C61
Suggested Citation: Suggested Citation