Multiclass Ensemble of One-Against-All SVM Classifiers



Ensembles of classifiers have recently received a resounding interest due to their successful application in different scenarios. In this paper, our main focus is on using ensembles of one-against-all classifiers in multiclass problems. Current approaches in multiclass problems are often focused in dividing the problem but seldom focus on cooperating strategies between classifiers.
We propose a framework based on Support Vector Machines (SVM) one-against-all baseline ensemble classifiers that includes a Multiclass Ensemble Function (MEF) to heuristically incorporate both the predictions of individual classifiers as well as the confidence margin associated with those predictions to determine the final ensemble output.
The results achieved with the renown Iris and Wine datasets show the performance improvement achieved by the proposed multiclass ensemble of one-against-all SVM classifiers.


Ensembles, Multiclass Classification, SVM


ISNN 2016 - International Symposium on Neural Networks, LNCS, Springer, S Petersburg, Russia, July 2016

Cited by

No citations found