HYBRID ARTIFICIAL INTELLIGENCE ARCHITECTURES FOR PREDICTING BEHAVIORAL ANOMALIES IN NETWORK TRAFFIC

Abstract

The relevance of this study is driven by the rapid growth in the volume and diversity of network traffic, increasing complexity of modern distributed network architectures, and a rising frequency of sophisticated cyber incidents manifested as behavioral anomalies. Under these conditions, traditional signature-based and isolated model-driven traffic analysis approaches fail to provide sufficient accuracy, robustness, and timeliness of response, which necessitates the adoption of integrated intelligent methods for predicting anomalous network behavior. The purpose of the article is to substantiate methodological approaches and to improve the effectiveness of predicting behavioral anomalies in network traffic through the use of hybrid artificial intelligence architectures, hereinafter referred to as AI, capable of providing adaptive, robust, and anticipatory detection of anomalous network behavior in dynamic and heterogeneous network environments. The research methods are based on a theoretical analysis of contemporary scientific sources in the fields of AI and cybersecurity, a systems approach to the study of network processes, logical generalization of results, and comparative analysis of approaches to anomaly prediction in network traffic using different classes of intelligent models. The research results demonstrate that the isolated application of individual AI models is of limited effectiveness for early prediction of behavioral anomalies, whereas their architectural integration enables comprehensive consideration of temporal, structural, and contextual characteristics of traffic. It is established that hybrid AI architectures improve prediction accuracy and robustness, reduce model sensitivity to data drift, and create a temporal advantage for preventive response. Key scientific and practical challenges in the implementation of such solutions are identified, including limited quality of training data, system scalability, interpretability of results, and risks of targeted model manipulation. Conclusions. Hybrid AI architectures should be considered a systemic foundation for intelligent network traffic analysis, enabling a transition from reactive incident detection to proactive, prediction-oriented cybersecurity management. Prospects for further research are associated with deepening the theoretical foundations of hybrid AI architecture design, developing harmonized criteria for evaluating the quality of early prediction, and investigating approaches to enhancing the robustness of intelligent traffic analysis systems in distributed and critical network infrastructures.