A similarity measure in Bayesian classification based on characteristic attributes of objects

Bayesian Networks provide an established approach to classification of objects in various types of activities. These range from technical via medical to economical applications. Heterogeneous sources provide observed declarations, which are evaluated by calculating the probabilities of resulting classes based on an underlying Bayesian Network model. Applying this classification process at different times, different sets of declarations of the same object can be observed. A similarity measure for uncertainty representation in a Bayesian Classification Network based on characteristic attributes of objects is proposed in order to determine the probability of two given sets of declarations originating from the same object. Based on a simulation, performance of this similarity measure is evaluated, given deferred sets of declarations from the same and from different objects. Possible applications include support of objects' re-recognition and track-to-track association by not necessarily positional or kinematical sources.