MULTICRITERIA OPTIMIZATION AND MANAGEMENT OF QUALITY INDICATORS IN MACHINE LEARNING PROBLEMS

Borzykh Nikita Yurievich, Smolentseva Tatyana Evgenievna

The article discusses the issues of multicriteria optimization and adaptive control of a set of quality criteria in machine learning problems. The limitations of single-criteria optimization are analyzed and the effectiveness of taking into account a set of indicators is demonstrated. Multicriteria optimization methods based on Pareto optimality and other approaches are proposed. Special attention is paid to the adaptive selection of the optimal set of criteria during the learning process using meta-learning based on recurrent neural networks. A network architecture is presented that allows dynamic selection of effective combinations of criteria. Examples of successful application of the developed approaches in predictive analytics problems and intelligent decision support systems are considered. The prospects for further research in the field of adaptive methods for multicriteria optimization of machine learning models are shown. Based on the results of the analysis, a meta-learning scheme for adaptive selection of criteria is proposed.

Keywords:machine learning, multicriteria optimization, meta-learning, decision support system, Pareto optimization