| Sumario: | The early detection of cyberattacks is critical to protecting enterprise networks. This paper proposes a method that uses data mining and machine learning techniques to identify harmful traffic on computer networks. The UNSW-NB15 dataset was used as a reference for testing this method. The CRISP-DM methodology was applied, ranging from understanding the data to evaluating the model. Univariate and bivariate exploratory analyses were carried out to select relevant characteristics for the study. Joint learning algorithms, such as Random Forest, Extra Trees, AdaBoost, and XGBoost, were used. Results show that models using the bagging method, particularly Random Forest, perform much better than boosting-based models in metrics such as accuracy (0.98), recall (0.99), and F1-score (0.98) in the attack category. It is concluded that collective learning approaches are an effective, understandable, and low-computational-cost alternative for automatically detecting intrusions in corporate networks. This study highlights the feasibility of integrating robust data analytics approaches into advanced cybersecurity systems.
|
|---|