In the literature, the introduction of the reject option inMultiple Classifier Systems has been analysed only fromthe experimental point of view. Following a firsttheoretical analysis provided by the authors, in thispaper, we analyse, in the framework of the minimum risktheory, the problem of finding the best error-reject tradeoffachievable by a linear combination of a given set oftrained classifiers. An algorithm for computing theparameters of the linear combination and of the rejectrule is then proposed. Experimental results on two datasets of remote-sensing images are reported.
A Method for Error Rejection in Multiple Classifier Systems
FUMERA, GIORGIO;ROLI, FABIO;
2001-01-01
Abstract
In the literature, the introduction of the reject option inMultiple Classifier Systems has been analysed only fromthe experimental point of view. Following a firsttheoretical analysis provided by the authors, in thispaper, we analyse, in the framework of the minimum risktheory, the problem of finding the best error-reject tradeoffachievable by a linear combination of a given set oftrained classifiers. An algorithm for computing theparameters of the linear combination and of the rejectrule is then proposed. Experimental results on two datasets of remote-sensing images are reported.
I metadati presenti in IRIS UNICA sono rilasciati con licenza Creative Commons CC0 1.0 Universal, mentre i file delle pubblicazioni sono protetti da diritto d'autore, salvo diversa indicazione.
