Evaluating the Effectiveness of Cyber Security Frameworks in Preventing Banking Fraud

Abstract

Banking fraud has become a significant challenge due to the rapid growth of digital financial transactions and evolving cyber threats. Cybersecurity frameworks play a crucial role in mitigating such risks by enhancing detection and prevention capabilities. This study aims to evaluate the effectiveness of different machine learning-based approaches for identifying fraudulent banking transactions within the context of cybersecurity controls. A synthetic yet realistic banking dataset consisting of 12,000 transactions from 3,000 customers was used for analysis. The dataset includes transaction details, behavioral attributes, and security-related features such as authentication methods and risk scores. Several classification models, including Logistic Regression, Random Forest, XGBoost, Support Vector Machine (SVM), and K-Nearest Neighbors (KNN), were implemented. The dataset was split into training (80%) and test (20%) sets, and performance was evaluated using accuracy, precision, recall, F1-score, ROC-AUC, and Precision-Recall metrics. The experimental results indicate that ensemble-based models such as Random Forest and XGBoost achieved higher accuracy (above 98%) and better overall performance compared to other models. However, despite high accuracy, all models struggled to detect minority fraud cases due to class imbalance, as reflected in lower recall and F1 Scores for the fraud class. Logistic Regression provided stable performance with interpretable results, while SVM and KNN showed comparatively lower effectiveness in fraud detection. The findings suggest that while cyber security-inspired feature engineering improves model performance, addressing class imbalance remains critical for reliable fraud detection. The study concludes that advanced ensemble techniques, combined with appropriate data-balancing strategies, can significantly enhance the effectiveness of cybersecurity frameworks in preventing banking fraud.