On a Lower Bound to Classification Error Probability in an Ensemble of Data Sources

For a given ensemble of data sources, we study a classification model in terms of error probability as a function of information quantity in a set of the submitted objects relative to their classes. An analytical lower bound to the average error probability depending on the average mutual information between the ensemble and a set of class-label decisions about the objects is found. For any collection of the discriminant functions, a redundancy of the average error probability relative to the lower bound is defined. Using some weak collections of the discriminant functions in the datasets of face and signature images as well as the compositions of these discriminant functions in the ensemble of the datasets, the comparative estimates of the error probability and the redundancy are calculated.