Comparison of identity fusion algorithms using estimations of confusion matrices

Scope of this paper is to investigate the performances of different identity declaration fusion algorithms in terms of probability of correct classification, supposing that the information for combination of the inferences from the different classifier is affected by measurement errors. In particular, these information have been assumed to be provided in the form of confusion matrices. Six identity fusion algorithms from literature with different complexity have been included in the comparison: heuristic methods such as voting and Borda Count, Bayes' and Dempster-Shafer's methods and the Proportional Redistribution Rule n° 1 in the Dempster-Shafer's framework.