Formation of recommendations for critical infrastructure objects based on streaming data, machine learning and artificial intelligence approaches

Abstract

The article addresses the problem of generating recommendations for critical infrastructure objects under conditions of increasing cyber threats, large volumes of streaming data, and the need for rapid decision-making. The relevance of the study is determined by the necessity to enhance the resilience of critical infrastructure through the use of intelligent decision support systems. The purpose of the research is to develop an integrated approach to recommendation generation based on streaming data, machine learning methods, Bayesian ranking, and explainable artificial intelligence. The study employs methods of anomaly detection, threat classification, risk forecasting, rule-based filtering, and XAI approaches for explaining generated recommendations. The proposed architecture provides real-time data processing and takes into account risks, security policies, and the operational context of the infrastructure object. Experimental validation on datasets demonstrated high system efficiency: F1 = 0.90, AUROC = 0.96, while processing latency did not exceed 0.8 s under a load of up to 10,000 messages per second. It was established that the adaptive self-updating mechanism reduces the number of false alarms by 43% and increases operators’ trust in the recommendation system. The obtained results confirm the prospects for the practical application of the proposed approach in supporting decision-making processes at critical infrastructure facilities.

Problems in programming 2026; 2: 16-27