Optimization of the partitions of a finite set with a nonadditive aggregati
on cost function (by a criterion similar to the criterion of "bottleneck" p
roblems, i.e., a minimax test) is studied. Two dynamic programming variants
, viz., a general optimization scheme for set functions and a dynamic progr
amming method for partitioning a "problem space" into intervals, are examin
ed. The clusters obtained in data processing are applied to a prediction pr
oblem.