Within the framework of multiple classifier fusion, linear combining plays an important role, because of its simplicity, its human understandability and its good theoretical basis. However in difficult tasks, linear combining of classifiers outputs can show unsatisfactory performance. In this paper we propose the use of a first-order Takagi-Sugeno-Kang (TSK) fuzzy model as improvement and extension of the linear combination rule. While the classical linear combining method assigns a weight to each pair (classifier, class), our approach is able to associate a weight with the triple (classifier, class, region of the classifier outputs space). In this way we can take the correlations between the classifier outputs into account. A technique to generate the TSK fuzzy model is also proposed. We performed a number of experiments using different classifiers on the Satimage and Phoneme data sets from the ELENA database. In almost all experiments, our combination method achieved better accuracy than the best single classifier. Further, we compared our model with 10 other techniques for classification fusion. We show that our method is in most cases superior (or substantially equivalent) to the other techniques on both data sets, and is still as simple to understand as the classical linear combining rule.

A TSK Fuzzy Model for Combining Outputs of Multiple Classifiers

COCOCCIONI, MARCO;LAZZERINI, BEATRICE;MARCELLONI, FRANCESCO
2004

Abstract

Within the framework of multiple classifier fusion, linear combining plays an important role, because of its simplicity, its human understandability and its good theoretical basis. However in difficult tasks, linear combining of classifiers outputs can show unsatisfactory performance. In this paper we propose the use of a first-order Takagi-Sugeno-Kang (TSK) fuzzy model as improvement and extension of the linear combination rule. While the classical linear combining method assigns a weight to each pair (classifier, class), our approach is able to associate a weight with the triple (classifier, class, region of the classifier outputs space). In this way we can take the correlations between the classifier outputs into account. A technique to generate the TSK fuzzy model is also proposed. We performed a number of experiments using different classifiers on the Satimage and Phoneme data sets from the ELENA database. In almost all experiments, our combination method achieved better accuracy than the best single classifier. Further, we compared our model with 10 other techniques for classification fusion. We show that our method is in most cases superior (or substantially equivalent) to the other techniques on both data sets, and is still as simple to understand as the classical linear combining rule.
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11568/206016
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 7
  • ???jsp.display-item.citation.isi??? 6
social impact