Approximations to Bayesian clustering rules

The amount of computation required for implementing the Bayesian cluster analysis suggested by Binder (1978) is often too large for exact results to be feasible. A general algorithm is proposed for approximating the similarity matrix and the resulting optimal partition. This algorithm is applied to artificial and to real data. For the real data, it appears that the algorithm is successful at identifying the optimal partitions as well as those units whose group membership is doubtful.