Abstract
Classifier ensemble techniques have been an active area of machine learning research in recent years. The aim of combining classifier ensembles is to improve the accuracy of the ensemble compared to any individual classifier. An ensemble can overcome the weakness of an individual classifier if its base classifiers do not make simultaneous errors. In this study, a novel algorithm for optimal classifier ensemble design called Coalition-based Ensemble Design (CED) is proposed and studied in detail. The CED algorithm aims to reduce the size and the generalization error of a classifier ensemble while improving accuracy. The underlying theory is based on the formation of coalitions in cooperative game theory. The algorithm estimates the diversity of an ensemble using the Kappa Cohen measure for multi base classifiers and selects a coalition based on their contributions to overall diversity. The CED algorithm is compared empirically with several classical design methods, namely Classic ensemble, Clustering, Thinning and Most Diverse algorithms. Experimental results show that the CED algorithm is superior in creating the most diverse and accurate classifier ensembles.
Highlights
Classifier ensembles have attracted the attention of many researchers as they can in crease the accuracy of classification for many tasks, especially the complex ones (Banfield et al, 2003)
The most important two factors when designing classifier ensembles are: how to select individual classifiers that are as diverse as possible and how to combine the different out puts in a way that enhances the ensemble accuracy (Alzubi, 2015). In literature it has been shown theoretically and proved through numerical result that classifier ensembles are efficient if and only if the base classifiers are of relatively high accuracy and don’t make simultaneous errors (Giacinto and Roli 2001a)
A construction of multi version systems of artificial neural networks by calculating a diversity measure was discussed by Partridge and Yates (1995)
Summary
Classifier ensembles have attracted the attention of many researchers as they can in crease the accuracy of classification for many tasks, especially the complex ones (Banfield et al, 2003). Our proposed approach is different from the above methods as it calculates the diversity for the entire ensemble and computes the diversity contribution of each individual base classifier which makes it possible to select only the classifiers with the most contribution. Other algorithms consider all classifiers in the ensemble and calculate one value for diversity.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callMore From: Research Journal of Applied Sciences, Engineering and Technology
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
