Comparative Study of Real-Time and Batch Processing Approaches in Machine Learning-Based Fraud Detection for Financial Institutions

This research investigates the comparative effectiveness of real-time and batch processing approaches in machine learning-driven fraud detection within financial institutions. Fraud detection in finance is a critical and evolving challenge, as fraudulent activities exploit system vulnerabilities and often appear as anomalies within a vast set of legitimate transactions. By implementing classification models K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Logistic Regression (LR) and deep learning models, including Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks, this study aims to differentiate and optimize fraud detection performance in both real-time and batch contexts. Real-time processing allows immediate analysis as transactions occur, enabling prompt fraud detection and response, which is essential in scenarios where instantaneous action is needed to prevent losses. Adopting Cross-Industry Standard Processing for Data Mining (CRISP-DM) Methodology leveraging on Google Colab and Python tools. Conversely, batch processing evaluates transactions collectively after a designated interval, providing a more comprehensive analysis by identifying patterns in larger datasets but at the expense of delayed detection. In this study, fraud is characterized by patterns such as unusually high transaction volumes, atypical geographic locations, and irregular transaction timings. Machine learning techniques are employed to analyze these features, distinguishing legitimate from potentially fraudulent transactions. Results indicate strengths and limitations in both processing modes, with real-time offering speed but potential data noise and batch providing accuracy yet delayed detection. This research underscores the value of a tailored approach, integrating machine learning models to enhance fraud detection efficacy and highlights the implications of processing choices for financial institutions aiming to strengthen security frameworks against evolving fraudulent strategies. Keywords: Anomaly Detection, Batch Processing, Convolutional Neural Networks (CNN), Fraud Detection, Financial Institutions, K-Nearest Neighbors (KNN), Logistic Regression (LR), Long Short-Term Memory (LSTM), Machine Learning, Real-Time Processing.