Nelson, B. Biggio & P. Laskov; JMLR W&CP 20:63–79, 2011.
Microbagging Estimators: An Ensemble Approach to Distance-weighted Classiﬁers
Support vector machines (SVMs) have been the predominate approach to
kernel-based classiﬁcation. While SVMs have demonstrated excellent performance in many
application domains, they are known to be sensitive to noise in their training dataset.
Motivated by the equalizing eﬀect of bagging classiﬁers, we present a novel approach to
kernel-based classiﬁcation that we call microbagging. This method bags all possible
maximal-margin estimators between pairs of training points to create a novel linear kernel
classiﬁer with weights deﬁned directly as functions of the pairwise distance matrix
induced by the kernel function. We derive relationships between linear and distance-based
classiﬁers and empirically compare microbagging to the SVMs and robust SVMs on several
Page last modified on Sun Nov 6 15:42:25 2011.