Author (s) :

---

1). Abraham Danlami, Federal University, Wukari, Taraba State, Nigeria
2). E. J. Garba, Modibbo Adama University, Yola, Adamawa State, Nigeria
3). Y. M. Malgwi, Modibbo Adama University, Yola, Adamawa State, Nigeria
4). S. E. Dogo, Taraba State University, Jalingo, Taraba State, Nigeria

Abstract :

---

A bank network is an interconnected system designed to facilitate financial transactions, minimize customer waiting times, and reduce the risk of errors caused by automated systems or bots. However, over the past three decades, cyberattacks have emerged as a significant threat to the security of these networks. The tools like penetration testing, machine learning classifiers, multi-factor authentication, network traffic analysis, and bot detection systems are reactive rather than proactive. As a result, these measures may fail to detect or prevent sophisticated attacks in real time. Such vulnerabilities can be exploited by attackers to gain unauthorized access to banking networks, posing serious risks to data integrity and customer privacy. In this study, we aim to enhance the performance of an anomaly-based Intrusion Detection System (IDS) within a banking environment by developing a Stacked Hybrid Classifier. To achieve this, the study employed the CICIDS2017 dataset, which contains realistic traffic data simulating various types of attacks and benign behaviors. The dataset's inherent class imbalance common in intrusion detection scenarios was addressed using the Synthetic Minority Oversampling Technique (SMOTE), ensuring more balanced training and improved sensitivity to minority attack classes. The model was implemented using Python, stacking multiple base classifiers including Random Forest, Decision Tree, Logistic Regression, and K-Nearest Neighbors. The predictions from these models were combined to form a meta-model, resulting in a more robust and generalized detection capability. The study was evaluated using stratified k-fold cross-validation to ensure consistent and unbiased performance assessment across different data partitions. The results show that the stacked Hybrid classifier achieved the best accuracy of 99.76%, along with superior precision, recall, and F1 scores when compared to the individual classifiers. This demonstrates the effectiveness of Hybrid learning, data balancing, and rigorous validation in improving IDS performance in banking networks.

---