Improving Threat Detection in Information Security with Ensemble Learning

In the face of increasingly complex and frequent cyberattacks, traditional rule-based threat detection systems often fail to identify evolving malicious behaviours. This study addresses the challenge by leveraging ensemble learning to enhance intrusion detection in information security. By integrating three distinct machine learning models – Support Vector Machine (SVM), Random Forest, and Deep Neural Network (DNN) – the proposed approach capitalises on their strengths while mitigating their weaknesses. The primary goal is to enhance detection accuracy, minimise false positives, and ensure reliable performance across various attack types. Using benchmark datasets, such as NSL-KDD and CICIDS2017, each model is trained and evaluated separately before being combined through a voting mechanism. Results from 10-fold cross-validation show that while baseline models perform well individually, the ensemble demonstrates more balanced and robust detection, achieving 94.00% accuracy, 95.10% precision, and a high area under the curve score of 0.77. These findings highlight the value of ensemble methods in producing consistent and dependable threat classification. The contribution of this work lies in demonstrating how a multi-model ensemble strategy can significantly strengthen cybersecurity defences, offering a scalable solution adaptable to real-world security environments.