Forecasting the financial sustainability of critical infrastructure enterprises based on cloud computing

Abstract

The study aims to work out a reproducible forecasting model to determine the financial sustainability of Ukrainian critical infrastructure businesses during the war. The analysis is performed on the basis of a balanced panel of 72 enterprises on 4 years 2019-2024 in the energy, transport, telecommunications, and water industries, incorporating econometric models (probit, cox survival analysis), machine learning models (Random Forest, XGBoost, and LSTM), and a hybrid ensemble specification. Models are implemented and deployed in a Microsoft Azure cloud computing environment to ensure scalability and security, and enable real-time forecasting. Results indicate that liquidity limitations, leverage, and operational disruption are the leading causes of financial distress, as the incidence of distress has increased by 31% since 2022, compared to 14% during the pre-war period. In terms of predictive performance, XGBoost achieves an out-of-sample AUC of 0.89. At the same time, the hybrid ensemble model outperforms all individual specifications, with an AUC of 0.92, an accuracy of 0.85, and an RMSE of 0.28. Results show that econometric interpretability, with machine-learning predictive power, is significantly better at early warnings. In practice, the framework offers Ukrainian policymakers and infrastructure managers a scalable tool of active risk surveillance, special financial aid, and resilience-oriented decision-making during the recovery.