Choice of optimal complexity of the class of logical decision functions in pattern recognition problems

A Bayesian recognition model based on a finite set of events has been proposed that can be used to develop and analyze methods for constructing logical decision functions. The model, when choosing an optimal complexity of the class of the logical decision functions in pattern recognition problems, takes into consideration empirical data and expert knowledge. The Bayesian pattern recognition model is defined by formulating certain statements that avoid the local metric properties of the feature space. For determining an optimal logical decision functions in pattern recognition problems, the space of variables is partitioned into a sufficiently large number of subdomains. The original set of partition subdomains or its extension obtained by uniting the subdomains is treated as a finite set of events. The various versions of partitions are searched using a certain algorithm and the version that is optimal according to the given criterion is returned as a decision.