NMC: nearest matrix classification - A new combination model for pruning One-vs-One ensembles by transforming the aggregation problem

详细信息    查看全文

• 作者：Mikel Galar ; ^a ; ^e ; ^{mikel.galar@unavarra.es} ; Alberto Ferná ; ndez^b ; ^{alberto.fernandez@ujaen.es} ; Edurne Barrenechea^a ; ^e ; ^{edurne.barrenechea@unavarra.es} ; Humberto Bustince^a ; ^e ; ^{bustince@unavarra.es} ; Francisco Herrera^c ; ^d ; ^{herrera@decsai.ugr.es}
• 关键词：Multi-class classification ; One-vs-One ; Decomposition strategies ; Classifier selection ; Ensemble pruning
• 刊名：Information Fusion
• 出版年：2017
• 出版时间：July 2017
• 年：2017
• 卷：36
• 期：Complete
• 页码：26-51
• 全文大小：1092 K
• 卷排序：36

文摘

A new combination strategy for OVO is proposed by transforming the aggregation problem. New instances are classified by the similarity of their outputs with respect to those of the training instances. The possibility of carrying out pruning in OVO ensembles is introduced for the first time. An exhaustive experimental study showing the existence of redundant (non-necessary) classifiers in OVO is developed.