A MACHINE LEARNING FRAMEWORK FOR FRAUD DETECTION AND SUSPECT PROFILING IN THE BANKING SECTOR

ABSTRACT

The banking sector, which is central to the economic structure, has become a primary target for cybercrime due to the adoption of digital banking and the increasing sophistication of cybercriminals. This not only poses a danger to banking sectors and their clientele, but also to people’s livelihoods and the stability of the global economy. In addition, most banks do not have fraud detection systems that feature real-time action, adaptive mechanisms to learn during the investigation, or deep learning capabilities. As a result of this, there are many cases of fraud which are not acted upon in a timely manner or are completely ignored. This research proposes a framework based on machine learning that provides suspect profiling and fraud detection. We employed ensemble models alongside comparison with Logistic Regression, XGBoost, and other algorithms by utilizing the “Credit card fraud” dataset from Kaggle. To enhance model performance, feature extraction methods such as Principal Component Analysis (PCA) alongside correlation matrix analysis, were implemented. Additionally, the model incorporates real-time monitoring that ensures proactive fraud surveillance and pattern-based profiling actions. The results demonstrate significant predictive accuracy with a Random Forest recall rate of 0.77, a precision mark of 0.97 and a remarkable accuracy score of 99.96%. The design is adaptable and flexible, catering to future integrations into the core banking system. This effort extends beyond detecting fraudulent activity, as it provides a new approach to tracing and profiling such accounts, and remains relevant in both research and practice.

Keywords: Banking Sector, Fraud Detection, Machine Learning, Real-time Monitoring, Suspect Profiling.