Detecção de ameaças cibernéticas em redes de computadores : uma abordagem baseada em métodos não supervisionados
Shranjeno v:
| Glavni avtor: | |
|---|---|
| Publication Date: | 2025 |
| Format: | Bachelor thesis |
| Jezik: | por |
| Source: | Repositório Institucional da UPF |
| Download full: | https://repositorio.upf.br/handle/123456789/9866 |
Izvleček: | This study investigates the application of unsupervised machine learning methods for detecting cyber threats. The work stems from the growing challenge of identifying anomalous activities within large volumes of network traffic, in a context where digital interconnection continually expands the attack surface. Two widely used anomaly detection algorithms were evaluated: Isolation Forest and Local Outlier Factor. The CICIDS2017 dataset was employed, providing a comprehensive scenario of normal and malicious traffic and enabling the construction of a realistic experimental environment. After the data preprocessing stage, the models were applied both individually and in combination, allowing performance analysis in terms of accuracy, false positive rates and the ability to distinguish behavioral patterns. The results showed that the Local Outlier Factor achieved superior performance compared to Isolation Forest, particularly in reducing false positives. The hybrid approach, however, demonstrated the best balance between accuracy and consistency across results. The study concludes that combining techniques can enhance threat identification, contributing to the development of more robust network security systems. |
Komentirajte kot prvi!