par Elouedi, Zied;Mellouli, Khaled;Smets, Philippe 
Référence Lecture notes in computer science, 2143, page (350-361)
Publication Publié, 2001
Référence Lecture notes in computer science, 2143, page (350-361)
Publication Publié, 2001
Article révisé par les pairs
| Résumé : | We develop a method to evaluate the reliability of a sensor in a classification task when the uncertainty is represented by belief functions as understood in the transferable belief model. This reliability is represented by a discounting factor that minimizes the distance between the pignistic probabilities computed from the dis- counted beliefs and the actual values of the data in a learning set. We then describe a method to tune the discounting factors of several sensors when their reports are merged in order to reach an aggregated report. They are computed so that together they minimize the distance between the pignistic probabilities computed from the combined dis- counted belief functions and the actual values of the data in a learning set. The first method produces the reliability of a sensor considered alone. The second method considers a set of sensors, and weights each of them so that together they produce the best predictor. |
