Integrated decision support system for cyber incident cost prediction in critical infrastructure

Authors

DOI:

https://doi.org/10.33216/1998-7927-2026-305-6-43-52

Keywords:

decision support system, critical infrastructure, time series forecasting, ensemble forecasting, Bayesian aggregation, uncertainty estimation

Abstract

Critical infrastructure systems operate under conditions of high dynamics, non-stationarity, and uncertainty, which imposes stringent requirements on modern decision support systems. Timely and reliable forecasting of the condition of such facilities is an essential prerequisite for improving their reliability and operational safety. Existing forecasting methods provide high predictive accuracy; however, they insufficiently account for prediction uncertainty, offer limited interpretability of results, and exhibit inadequate adaptability to changing operating conditions, while forecasting, uncertainty estimation, and explainability are mostly implemented in isolation from one another. The aim of this study is to improve the reliability, adaptability, and interpretability of decision support by developing an integrated system for forecasting the condition of critical infrastructure facilities.

A modular system architecture is proposed that integrates data acquisition and preprocessing, feature engineering, ensemble forecasting, uncertainty estimation, explainability, and decision support modules. The hybrid ensemble combines statistical models, machine learning regressors, and recurrent neural networks with dynamic Bayesian weighting of their forecasts. The scientific novelty lies in the integration of hybrid forecasting, dynamic Bayesian model aggregation, prediction uncertainty estimation, and adaptive learning with a closed-loop feedback mechanism within a unified decision support information technology.

The experimental evaluation was carried out on the open ETTh1 dataset (17,420 observations, 41 features) using a walk-forward protocol (20 evaluation windows, forecast horizons of 1–24 hours). The proposed system achieved MAE = 1.329 ± 0.448, RMSE = 1.911 and R² = 0.897, which corresponds to an error reduction of 24.6–30.1 % compared with individual machine learning models. The ablation study showed that the adaptive learning module accounts for an 82.4 % reduction in MAE, while the Bayesian aggregation module ensures correct calibration of prediction intervals (PICP = 0.952 at a nominal level of 0.95). The confidence index correlates significantly with the actual forecast error, which allows it to be used as a criterion for escalating decisions to a human expert. The statistical significance of the differences was confirmed by the Wilcoxon and Friedman tests.

References

1. Hyndman R. J., Athanasopoulos G. Forecasting: Principles and Practice. 3rd ed. Melbourne : OTexts, 2021. 442 p. URL: https://otexts.com/fpp3/

2. Tao Hong, Pierre Pinson, Shu Fan, Global Energy Forecasting Competition 2012, International Journal of Forecasting, Volume 30, Issue 2, 2014, 357-363, https://doi.org/10.1016/j.ijforecast.2013.07.001

3. Box G. E. P., Jenkins G. M., Reinsel G. C. Time Series Analysis: Forecasting and Control. 4th ed. Hoboken : John Wiley & Sons, 2008. 784 p. https://doi.org/10.1002/9781118619193

4. Hochreiter S., Schmidhuber J. Long short-term memory // Neural Computation. 1997. Vol. 9, No. 8. P. 1735–1780. DOI: https://doi.org/10.1162/neco.1997.9.8.1735

5. Cho K., van Merrienboer B., Gulcehre C. et al. Learning phrase representations using RNN encoder–decoder for statistical machine translation // Proc. EMNLP. 2014. P. 1724–1734. DOI: https://doi.org/10.3115/v1/D14-1179

6. Gal Y., Ghahramani Z. Dropout as a Bayesian approximation: representing model uncertainty in deep learning // Proc. ICML. 2016. P. 1050–1059. DOI: https://doi.org/10.48550/arXiv.1506.02142

7. Zhou Z.-H. Ensemble Methods: Foundations and Algorithms. Boca Raton : CRC Press, 2012. 236 p. DOI: https://doi.org/10.1201/b12207

8. Hoeting J. A., Madigan D., Raftery A. E., Volinsky C. T. Bayesian model averaging: a tutorial // Statistical Science. 1999. Vol. 14, No. 4. P. 382–401. DOI: https://doi.org/10.1214/ss/1009212519

9. Raftery A. E., Madigan D., Hoeting J. A. Bayesian model averaging for linear regression models // Journal of the American Statistical Association. 1997. Vol. 92, No. 437. P. 179–191. DOI: https://doi.org/10.1080/01621459.1997.10473615

10. Lundberg S. M., Lee S.-I. A unified approach to interpreting model predictions // Advances in Neural Information Processing Systems. 2017. Vol. 30. P. 4765–4774. DOI: https://doi.org/10.48550/arXiv.1705.07874

11. Ribeiro M. T., Singh S., Guestrin C. "Why should I trust you?" Explaining the predictions of any classifier // Proc. 22nd ACM SIGKDD. 2016. P. 1135–1144. DOI: https://doi.org/10.1145/2939672.2939778

12. Barredo Arrieta A., Diaz-Rodriguez N., Del Ser J. et al. Explainable artificial intelligence (XAI): concepts, taxonomies, opportunities and challenges toward responsible AI // Information Fusion. 2020. Vol. 58. P. 82–115. DOI: https://doi.org/10.1016/j.inffus.2019.12.012

13. Gama J., Zliobaite I., Bifet A., Pechenizkiy M., Bouchachia A. A survey on concept drift adaptation // ACM Computing Surveys. 2014. Vol. 46, No. 4. Article 44. DOI: https://doi.org/10.1145/2523813

14. Mosqueira-Rey, E., Hernández-Pereira, E., Alonso-Ríos, D. et al. Human-in-the-loop machine learning: a state of the art. Artif Intell Rev 56, 3005–3054 (2023). https://doi.org/10.1007/s10462-022-10246-w

Published

2026-09-29