Automation of Determining Criterion Weights in Decision Support Tasks Based on Regression Analysis Results

Currently, when operating the decision support system of a critical infrastructure facility’s situation center, difficulties arise in determining the weights of criteria for selecting the optimal solution. The main source of errors is expert weighting errors, due to the fact that different experts may assign significantly different weights to the same set of criteria. The paper proposes and experimentally investigates an approach to determining the weights of criteria in multi-criteria decision support tasks based on the use of machine learning methods and generative language models. Unlike classical analytical and expert methods, the proposed methodology formalizes the task of assigning weights as a regression task, in which each criterion is described by a set of statistical characteristics of the distribution of its estimates across a set of alternatives. The experiments conducted on the LLM-generated synthetic benchmark showed that the ML approach, in particular the Ridge regression model, achieved the lowest test MSE among the considered regression models and outperformed the analytical and combined baselines in the reported setting. These results suggest that regression-based estimation of synthetic reference criterion weights may be a promising direction for automating weight determination in MCDM tasks. However, since the current empirical evidence is based on synthetic data from one selected domain, the results should not be interpreted as proof of general validity for real expert decision-making settings without further validation on real-world MCDM datasets.