Podrobná bibliografie
| Hlavní autor: |
Vieira, Bernardo Vivian |
| Datum vydání: |
2025 |
| Médium: |
Bachelor thesis
|
| Jazyk: |
por |
| Zdroj: |
Repositório Institucional da UPF |
| Download full: |
https://repositorio.upf.br/handle/123456789/9865
|
Shrnutí: |
The growing sophistication of cyberattacks, especially adversarial attacks against artificial intelligence (AI) models, is an emerging challenge to digital security. This work investigates the effectiveness of AI techniques in mitigating such attacks in Intrusion Detection Systems (IDS), through experiments using the CIC-IDS-2017 dataset (2.83 million samples). The methodology includes training three machine-learning models (Decision Tree, Random Forest, MLP), evaluating adversarial robustness (FGSM, PGD, C&W), and measuring performance with metrics such as accuracy, detection rate, false-positive rate, and latency. Results show Random Forest maintains 83% accuracy under attack against approximately 50% for other models (p < 0.0001), though severe class imbalance causes 96% false negatives for minority attacks, limiting production viability without mitigation strategies. The text provides empirical evidence on AI resilience in IDS and identifies critical limitations for real-world deployment. |
