ARTIFICIAL INTELLIGENCE METHODS FOR DETECTING ATTACKS AND THE WAYS OF IMPROVING DETECTION ACCURACY

The rapid evolution of cyber threats demands intelligent, adaptive detection mechanisms beyond traditional rule-based systems. This paper reviews artificial intelligence (AI) methods -- including deep learning (DL), transformer-based architectures, federated learning (FL), and explainable AI (XAI) - applied to network intrusion detection systems (IDS) between 2021 and 2026. Seven state-of-the-art models are evaluated across modern benchmark datasets (CIC-IoT2023, BoT-IoT, TON_IoT, UNSW-NB15), achieving detection accuracies between 98.70% and 99.40% with False Alarm Rates (FAR) as low as 0.60%. Six core accuracyimprovement strategies are systematically analyzed: feature selection, data augmentation, attention mechanisms, federated training, explainability integration, and hyperparameter optimization. The paper highlights that no single technique is sufficient -- hybrid architectures combining temporal modeling, privacy-preserving distributed training, and human-interpretable explanations consistently deliver the highest real-world performance. The study concludes with open challenges including adversarial robustness, regulatory compliance under the EU AI Act, and dataset representativeness.

Hali tarjima qilinmagan