Порівняльний аналіз ефективності GNN та RNN у задачах регресійного аналізу

Abstract

This paper considers the application of graph neural networks for solving regression analysis problems. Modern information systems increasingly operate with complex multidimensional data that contain structural relationships between observations. Traditional regression models and classical neural networks often cannot fully take into account such dependencies. Graph neural networks allow representing data as graphs where nodes correspond to objects or system states and edges represent relationships between them. This approach enables more accurate modeling of complex systems. The paper analyzes the principles of graph neural networks and their advantages for regression tasks. Special attention is paid to attention-based architectures that allow adaptive weighting of neighboring nodes. The proposed approach can improve prediction accuracy in engineering and technical systems