CYBER THREATS PREDICTON USING EXPERIENCE SHARING MODEL AND ENSEMBLE LEARNING ALGORITHM

Title

CYBER THREATS PREDICTON USING EXPERIENCE SHARING MODEL AND ENSEMBLE LEARNING ALGORITHM

Author

1. Abubakar Bello, Student, National Open University of Nigeria, Nigeria

Abstract

The increasing complexity of cyber threats, particularly in critical industries such as oil and gas, necessitates proactive predictive models for early detection and response. Traditional frameworks such as the Common Vulnerability Scoring System (CVSS) are reactive, often addressing vulnerabilities post-incident, thereby exposing organizations to operational and financial risks. This study proposes a novel hybrid framework combining an experience-sharing model with ensemble machine learning algorithms, including bagging and boosting techniques. Using structured datasets such as VERIS and CAPEC, machine learning classifiers—logistic regression, k-Nearest Neighbors, and regression trees—were employed and validated using k-fold cross-validation. The results revealed a 94% prediction accuracy and a 0.96 AUC-ROC score with bagging ensembles, outperforming conventional models by 12%. A case study focused on Nigeria’s oil and gas infrastructure validated the model’s sector-specific applicability. This study contributes to cybersecurity analytics by demonstrating (1) the efficacy of ensemble learning, (2) a validated experience-sharing paradigm, and (3) the development of dynamic cyber-risk metrics suited for modern threats. The proposed framework offers cost-effective and scalable solutions for proactive threat mitigation.

Keywords

Conclusion

This study demonstrates the applicability of ensemble machine learning models for predicting cybersecurity threats, with a focus on critical infrastructure such as the Nigerian oil and gas sector. Using structured datasets and cross-validated ensemble models, the research achieved high accuracy and reliability. Notably, Random Forest and Gradient Boosting models performed best across key evaluation metrics.

Key contributions include the development of a domain-specific cyber threat prediction model, integration of experience-sharing frameworks, and validation of ensemble methods for cyber-risk quantification. These outcomes are particularly relevant for sectors requiring preemptive resource allocation and security incident mitigation.

Future research should explore deep learning models, zero-day threat detection, and real-time deployment integration with SIEM platforms. Localized datasets and cross-organizational collaboration can further enhance the model's utility and adaptability.

Author Contrubution

A.B.: Conceptualization, Methodology (ensemble learning model), Writing – Original Draft. A.B.: Software (Python implementation), Data Curation (VERIS/CAPEC datasets), Formal Analysis. A.S.: Validation (k-fold cross-validation), Writing – Review & Editing. A.B.: Supervision, Project Administration

Funding

This research received no external funding

Conflict of Interest

The authors declare no conflict of interest

Data Sharing Statement

The datasets analyzed in this study—VERIS (Vocabulary for Event Recording and Incident Sharing) and CAPEC (Common Attack Pattern Enumeration and Classification)—are publicly available at their respective sources: VERIS Community Database and CAPEC MITRE Repository. The derived datasets and code used for ensemble learning analysis are available from the corresponding author upon reasonable request.

Software And Tools Use

This study was implemented using Python 3.8 with key libraries including Scikit-learn (v1.0) for ensemble learning algorithms (bagging/boosting), Pandas (v1.3) for data processing, and Matplotlib (v3.4) for visualization. The analysis was conducted in Jupyter Notebook and Google Colab environments. Anaconda (v2021.05) was used for package management

Acknowledgements

We thank the Petroleum Technology Development Fund (PTDF), Nigeria, for their institutional support. We also acknowledge the VERIS and CAPEC communities for providing open-access datasets critical to this research. Special gratitude to ACETEL at National Open University of Nigeria for their technical guidance and to the anonymous reviewers for their constructive feedback.