Classification with imprecise likelihoods: A comparison of TBM, random set and imprecise probability approach

The problem is target classification in the circumstances where the likelihood models are imprecise. The paper highlights the differences between three suitable solutions: the Transferrable Belief model (TBM), the random set approach and the imprecise probability approach. The random set approach produces identical results to those obtained using the TBM classifier, provided that equivalent measurement models are employed. Similar classification results are also obtained using the imprecise probability theory, although the latter is more general and provides more robust framework for reasoning under uncertainty.