MODELING AND EVALUATION OF THE PROPERTIES OF AN ADAPTIVE ENSEMBLE INDICATOR FOR DIAGNOSING THE STRATEGIC RESOURCE SECURITY OF A DISTRIBUTED ORGANIZATIONAL SYSTEM

Abstract

The paper presents the results of a study of the properties of an adaptive ensemble indicator for diagnosing the strategic resource security (DSRS) of a distributed organizational system, intended to support decision-making under conditions of high uncertainty, incomplete information, and class imbalance. The indicator is implemented as a heterogeneous ensemble of binary classifiers combining machine learning models of different paradigms (Naive Bayes, Support Vector Machine, Random Forest, k-Nearest Neighbors, Logistic Regression) and is supplemented with specialized mechanisms for feature scaling, probability calibration, and adaptive weighting of model outputs. The relevance of the study is determined by the specific nature of resource data in strategic management systems, which are characterized by complex and asymmetric feature distributions, temporal lags, high heterogeneity of observation conditions, dominance of satisfactory states, and the critical importance of minimizing the risk of hypodiagnosis of unfavorable situations. Taking these characteristics into account, the paper proposes a composite quality metric KQ that integrates the F1neg score, the Matthews correlation coefficient, and Cohen’s kappa, while considering their mutual correlations and domain-specific managerial priorities. The research is conducted through a series of computational experiments using specially generated synthetic datasets that reproduce domain-driven statistical properties of resource observations, including heavy-tailed distributions, label noise, severe class imbalance, and the influence of historical resource dynamics. For this purpose, a unified framework for managing computational experiments is developed, ensuring reproducibility, comparability of results, and statistical validation of observed effects. Within three experimental studies, the algorithm is subjected to baseline verification, a stepwise analysis of the contribution of individual mechanisms to classification quality, and an evaluation of ensemble performance under stress conditions in comparison with the best-performing individual classifier. The results demonstrate a stable advantage of the proposed ensemble DSRS indicator in terms of the composite quality metric, specificity, and F1neg score, which is confirmed by statistically significant paired tests and effect size estimates. It is shown that the proposed approach provides more reliable detection of potentially dangerous resource states and forms an interpretable basis for supporting strategic decision-making in distributed organizational systems.