Artificial Intelligence-based Models and Algorithms for Detecting Network Attack Traces in Computer Networks

Abstract

The increasing diversity and complexity of cyber threats have made the detection of network attack traces a critical task in modern computer networks. Conventional Intrusion Detection and Prevention Systems (IDS/IPS), particularly signature-based approaches, provide effective detection of previously known threats but remain limited when dealing with zero-day attacks, polymorphic malware, advanced persistent threats, and previously unseen malicious behavior. This study analyzes models and algorithms for detecting traces of network attacks, with particular emphasis on artificial intelligence-based approaches. Signature-based, anomaly-based, and hybrid detection mechanisms are examined, and their capabilities and limitations are considered in relation to evolving network threats. Particular attention is given to Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and Transformer architectures because of their ability to learn temporal, spatial, and contextual patterns from network traffic. The study further considers benchmark datasets, including CICIDS2017, UNSW-NB15, NSL-KDD, and TON_IoT, together with evaluation criteria such as accuracy, precision, recall, F1-score, and AUC-ROC for assessing intrusion detection performance. The comparative analysis indicates that conventional IDS/IPS approaches remain effective and computationally efficient for known attack signatures, whereas AI-based models provide greater adaptability for identifying complex and previously unseen patterns. However, their practical deployment introduces challenges related to computational requirements, model interpretability, data quality, model drift, scalability, and integration with existing security infrastructure. Based on these findings, a hybrid detection architecture combining deterministic signature-based mechanisms with adaptive AI-based analysis is identified as a promising direction for developing robust and scalable network attack detection systems.