Autor: Abreu Esttebam Tavares Quevedo (Currículo Lattes)
Resumo
Network traffic constitutes one of the primary means of communication today and is essential for the proper functioning of various everyday activities. In the globalized context of the internet, numerous malicious actors seek to cause harm or extort victims, with Distributed Denial of Service (DDoS) attacks representing a critical threat to network stability. Although several models have been proposed, they remain far from achieving optimal performance in modern infrastructures. This study aims to evaluate the impact of a Feature Engineering approach on enhancing the performance of DDoS prediction algorithms Random Forest and XGBoost. Specifically, the work proposes the optimization of predictive models by generating new features through an aggregation method based on the Generalized Choquet Integral with an adaptive alpha parameter. By applying this method to the most relevant features identified by the SelectKBest algorithm, the study aims to effectively model complex dependencies among network variables that conventional methods typically ignore. Experimental results show that incorporating these new fuzzy-based features enhances predictive models, allowing Random Forest and XGBoost algorithms to achieve higher accuracy and stability even with a reduced feature set.
Palavras-chave:
Anomaly detection, Choquet Integral generalized, Feature Engineering, DDoS Detection, Fuzzy Logic.