MODELS AND ALGORITHMS FOR DETECTING TRACES OF NETWORK ATTACKS IN COMPUTER NETWORKS

Abstract Since the threats in cyber space keep on increasing in number, variety and complexity, the detection of the traces of the attacks on the networks has become an exceptional concern of organizations in different parts of the globe. Classic Intrusion Detection and prevention systems (IDS/IPS) have their root cause in fixed focus, but are progressively inclined to the fluctuating attack vectors with zero-day exploits, advanced persistent threats (APTs) and exceptional malware. The proposed study explores contemporary networks and algorithms to identify traces of attacks on networks, and, in particular, how solutions incorporating artificial intelligence (AI) can be applied to this problem. The paper examines critically the performance of signature-based detection and anomaly-based detection and hybrid detection mechanisms and why the adoption of AI-driven approaches, like Long Short-Term Memory (LSTM) networks, Convolutional Neural Networks (CNNs), and Transformer architectures, can lead to improved accuracy, scalability, and responsiveness in real-time detection. Based on benchmark datasets and performance metrics, we perform the comparison of modern IDS/IPS structures and the classical ones. AI-based models. The results of our study indicate that AI models have a greater hand capturing the unidentified attack patterns and the complex patterns as compared to traditional systems which mostly run well on the known threats. The study will end with the proposed next steps in the creation of hybrid frameworks that avoid the weaknesses of deterministic rule-based systems and provide greater flexibility, similar to what AI can offer, and lead to more sustainable and future-proof network security plans.