Enhancement of the Heuristic Optimization
Based on Extended Space Forests using Classifier Ensembles
Zeynep Kilimci1,3
and Sevinç Omurca2
1Department of Computer Engineering, Dogus
University, Turkey
2Department of Computer Engineering, Kocaeli
University, Turkey
3Department of Information
Systems Engineering, Kocaeli University, Turkey
Abstract: Extended space forests are a matter of common knowledge for ensuring
improvements on classification problems. They provide richer feature space and
present better performance than the original feature space-based forests. Most
of the contemporary studies employs original features as well as various
combinations of them as input vectors for extended space forest approach. In
this study, we seek to boost the performance of classifier ensembles by
integrating them with heuristic optimization-based features. The contributions
of this paper are fivefold. First, richer feature space is developed by using
random combinations of input vectors and features picked out with ant colony
optimization method which have high importance and not have been associated
before. Second, we propose widely used classification algorithm which is
utilized baseline classifier. Third, three ensemble strategies, namely bagging,
random subspace, and random forests are proposed to ensure diversity. Fourth, a
wide range of comparative experiments are conducted on widely used biomedicine
datasets gathered from the University of California Irvine (UCI) machine
learning repository to contribute to the advancement of proposed study.
Finally, extended space forest approach with the proposed technique turns out
remarkable experimental results compared to the original version and various
extended versions of recent state-of-art studies.
Keywords: Classifier ensembles, extended space forests, ant
colony optimization, decision tree.
Received November 11,
2017; accepted March 11, 2018
https://doi.org/10.34028/iajit/17/2/6