Abstract

Due to the rapid growth of technology, the urgency for effective cybersecurity systems has become increasingly critical, notably within the paradigm of the Internet of Things (IoT) and Cloud services. With an increasing number of security features in communication protocols, there is a heightened need for systems to manage this complexity and protect assets efficiently. Machine learning (ML) techniques are increasingly indispensable in identifying and mitigating cyber threats. However, the vast number of security features can affect the performance of these techniques. This research presents a hybrid feature selection approach integrating Correlation Analysis (CA) and Mutual Information (MI) as filter methods. Moreover, Recursive Feature Elimination with Cross-Validation (RFECV) is also integrated as a wrapper method to identify highly ranked security features efficiently. These selected features are then deployed in tree-based classifiers, namely, Decision Tree (DT) and Random Forest (RF) classifiers, for predicting cyber-attacks. The proposed system is validated using a real-world dataset specific to a Network Intrusion Detection System (NIDS). The empirical results demonstrate that it can detect attacks effectively and significantly reduce the computational complexity compared to existing approaches. Therefore, the proposed system can enhance cybersecurity measures in complex network environments.