Enhancing Banking Cybersecurity: An Ensemble-Based Predictive Machine Learning Approach

Аннотация

In this study, we propose a predictive cybersecurity framework for the banking sector by integrating ensemble-based machine learning models. Our approach leverages heterogeneous datasets— including internal firewall and intrusion detection system logs, banking transaction records, user behavior data, and external threat intelligence—to capture a comprehensive view of the cyber threat landscape. Following rigorous data preprocessing, feature selection, and feature engineering, we evaluated multiple models, including Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, and Deep Neural Networks. Comparative analysis revealed that while advanced individual models demonstrated strong predictive capabilities, the Ensemble Model consistently outperformed all others, achieving an accuracy of 92% The American Journal of Engineering and Technology 88 https://www.theamericanjournals.com/index.php/tajet The American Journal of Engineering and Technology and a ROC-AUC of 94%. These results underscore the novel patterns indicative of emerging cyber-attacks. model’s superior ability to minimize false negatives, LITERATURE REVIEW which is critical for safeguarding financial assets. Our The application of machine learning in cybersecurity has findings advocate for the adoption of ensemble been a vibrant area of research over the past decade. techniques in real-world banking cybersecurity Early works in this domain primarily focused on anomaly applications, providing a robust, scalable solution that detection using statistical techniques and rule-based adapts to evolving threat patterns while significantly systems. For instance, Patcha and Park [1] provided an enhancing detection performance. early framework for intrusion detection systems (IDS) that relied on predefined rules and statistical anomalies. Keywords: Cybersecurity, Banking, Predictive However, as cyber threats became more sophisticated, Modeling, Ensemble Learning, Machine Learning, Data these traditional methods began to show limitations in Preprocessing, Feature Engineering, Threat Detection, their ability to detect novel attack patterns and adapt to ROC-AUC, Intrusion Detection. the evolving threat landscape. Subsequent research has shifted towards employing machine learning algorithms that can learn from data, Introduction: In today’s digital era, the banking sector thereby enhancing detection accuracy. Sommer and faces an unprecedented level of cyber threats that can Paxson [2] discussed the evolution of network intrusion compromise sensitive financial data and destabilize detection systems by integrating machine learning trust in financial institutions. Cybersecurity breaches techniques, highlighting how supervised and have evolved from simple malware infections to unsupervised learning can uncover previously complex, coordinated attacks that exploit undetected patterns. Various studies have vulnerabilities in network infrastructures and user demonstrated the efficacy of classifiers such as Support behaviors. In response, the integration of predictive Vector Machines (SVMs), Decision Trees, and Neural machine learning models has emerged as a promising Networks in predicting cybersecurity threats [3][4]. In strategy to enhance the security posture of banks. particular, Random Forests and Gradient Boosting These models offer the ability to analyze vast amounts Machines have been extensively used due to their of heterogeneous data—from network logs to user robustness in handling noisy and imbalanced data, as transactions—in near real-time, thereby identifying evidenced by the work of Chen et al. [5]. and mitigating potential threats before they materialize into significant breaches. Recent literature emphasizes the importance of ensemble methods, where the combination of multiple Our work addresses the pressing need for advanced machine learning models leads to improved predictive frameworks in cybersecurity by leveraging performance compared to individual models. Lippmann state-of-the-art machine learning techniques. In the et al. [6] illustrated that ensemble approaches, such as banking industry, where the repercussions of security bagging and boosting, are capable of reducing breaches are both financially and reputationally overfitting and enhancing generalization in IDS. devastating, traditional rule-based systems are no Similarly, a study by Javaid et al. [7] on advanced longer sufficient. Instead, predictive models capable of persistent threats in network environments learning from historical data and adapting to emerging demonstrated that ensemble models significantly threat patterns are critical for proactive defense outperform single classifier approaches in terms of mechanisms. This paper proposes an ensemble-based accuracy and robustness. approach that combines the strengths of multiple machines learning algorithms, ultimately providing a Moreover, the banking sector has specific challenges more robust, accurate, and adaptable syst