Robust Bayesian cluster enumeration based on the t distribution

A major challenge in cluster analysis is that the number of data clusters mostly unknown and it must be estimated prior to clustering observed data. In real-world applications, often subject heavy tailed noise outliers which obscure true underlying structure Consequently, estimating becomes challenging. To this end, we derive a robust enumeration criterion by formulating problem as maximization posterior probability multivariate tν distributed candidate models. We utilize Bayes’ theorem asymptotic approximations come up with possesses closed-form expression. Further, refine derivation provide for sets finite sample size. The criteria require an estimate parameters each model input. Hence, propose two-step algorithm uses expectation partition calculation one criteria. performance proposed tested compared existing methods using numerical real experiments.