Convex cluster stabilization of classification algorithms as a means for finding collective solutions with high generalization ability

A collective solution method in pattern recognition based on the simultaneous improvement of the stability and efficiency (the percentage of correctly classified objects in the learning sample) is generalized. The relationship between the procedure described in the paper and several available methods for constructing collective algorithms that are particular cases of a more general approach is revealed. The practical value of the method is confirmed by solving some well-known classification problems.