Architecture of a software system for multi-model ecological risk forecasting for digital management of municipal organic waste

• Дата: 2026
• Автори: Tryhuba, A.M., Koval, N.Ya., Tryhuba, I.L., Firman, I.R., Famuliak, V.Yu.
• Формат: Стаття
• Мова: Українська
• Опубліковано: PROBLEMS IN PROGRAMMING 2026
• Теми: ecological risk; multi-model forecasting; municipal organic waste; machine learning; Random Forest; Gradient Boosting; XGBoost; UML architecture; permutation importance; digital management; information system; decision support; UDC 004.94:628.4:504.064
• Онлайн доступ: https://pp.isofts.kiev.ua/index.php/ojs1/article/view/889
• Назва журналу: Problems in programming

The article investigates the problem of designing a software system for multi-model ecological risk forecasting for digital management of municipal organic waste under conditions of increasing volumes of significant data and the need for prompt decision-making at the local level. An analysis of modern approaches to ecological risk forecasting in municipal management systems has been carried out. It has been concluded that the use of individual models does not ensure sufficient stability of results under conditions of heterogeneous input data and nonlinear influence factors. The expediency of constructing a software system based on a multi-level architecture has been substantiated, including modules for data preparation, predictive analysis, model evaluation, feature interpretation, and digital decision support. A multi-model algorithm has been proposed, incorporating Random Forest, Gradient Boosting, and XGBoost, which makes it possible to automatically determine the most effective model according to the criteria MAE, RMSE, and R². A UML architecture of interaction between the classes DataManager, RiskModelEngine, BenchmarkController, and DecisionSupport has been developed. To explain the results, the permutation feature importance method has been applied, which made it possible to determine the dominance of spatial and infrastructural factors in risk formation. Experimental testing results showed that the highest accuracy is provided by the XGBoost model (MAE=0.039, RMSE=0.055, R²=0.930), confirming the effectiveness of the proposed approach.Problems in programming 2026; 1: 25-39