- Oche Akiti Ojoje1, Gilbert I.O. Aimufua2, Steven Ita Bassey3, Umaru Musa4
- DOI: 10.5281/zenodo.17751150
- SSR Journal of Engineering and Technology (SSRJET)
In an era of pervasive digital
connectivity, cyber-attacks have become increasingly sophisticated, persistent,
and difficult to detect using conventional security mechanisms. Traditional
intrusion detection systems (IDS) often rely on single classifier models, which
tend to underperform when faced with complex, dynamic, and high-dimensional
network data. This research proposes an ensemble-based predictive model for
cyber attack detection that integrates multiple machine learning algorithms to
enhance detection accuracy, robustness, and generalization. The model employs a
hybrid ensemble strategy combining bagging and boosting techniques, utilizing
algorithms such as Random Forest, Gradient Boosting, and Support Vector
Machines (SVM) to leverage the strengths of diverse learners while minimizing
their individual weaknesses.
The study utilizes benchmark
cybersecurity datasets such as NSL-KDD and CICIDS2017, which encompass a wide
range of network intrusions including Denial-of-Service (DoS), Probe, R2L, and
U2R attacks. Data preprocessing techniques—comprising feature encoding,
normalization, and dimensionality reduction—are applied to ensure optimal
learning conditions and minimize noise interference. The ensemble model is
trained and evaluated using performance metrics including accuracy, precision,
recall, F1-score, false positive rate (FPR), and ROC-AUC to measure both
detection efficiency and model reliability.
Experimental results demonstrate that
the ensemble-based model significantly outperforms individual classifiers in
identifying both known and zero-day attacks. The proposed system achieves high
detection accuracy while maintaining a low false positive rate, which is
critical for real-world cybersecurity applications. The hybrid ensemble approach
proves effective in addressing data imbalance, model overfitting, and
classification bias commonly associated with standalone models. Moreover, the
evaluation results indicate that ensemble learning enhances decision stability
and adaptability in detecting evolving attack patterns.
The research concludes that ensemble-based predictive modelling offers a scalable, reliable, and intelligent framework for next-generation intrusion detection systems (IDS). The findings underscore the importance of integrating multiple learning paradigms to build resilient cybersecurity infrastructures. Future research is recommended to explore deep ensemble learning, real-time adaptive learning systems, and integration with cloud-based security architectures to further improve predictive performance and operational scalability in dynamic network environments.

