We propose several novel methods for enhancing the multi-class SVMs byapplying the generalization performance of binary classifiers as the core idea.This concept will be applied on the existing algorithms, i.e., the DecisionDirected Acyclic Graph (DDAG), the Adaptive Directed Acyclic Graphs (ADAG), andMax Wins. Although in the previous approaches there have been many attempts touse some information such as the margin size and the number of support vectorsas performance estimators for binary SVMs, they may not accurately reflect theactual performance of the binary SVMs. We show that the generalization abilityevaluated via a cross-validation mechanism is more suitable to directly extractthe actual performance of binary SVMs. Our methods are built around thisperformance measure, and each of them is crafted to overcome the weakness ofthe previous algorithm. The proposed methods include the Reordering AdaptiveDirected Acyclic Graph (RADAG), Strong Elimination of the classifiers (SE),Weak Elimination of the classifiers (WE), and Voting based Candidate Filtering(VCF). Experimental results demonstrate that our methods give significantlyhigher accuracy than all of the traditional ones. Especially, WE providessignificantly superior results compared to Max Wins which is recognized as thestate of the art algorithm in terms of both accuracy and classification speedwith two times faster in average.
展开▼
