Artificial Intelligence-Driven Business Analytics for Real-Time Fraud Detection in Financial Organizations

Abstract

Financial organizations require intelligent fraud-detection systems capable of identifying suspicious transactions accurately and supporting rapid operational decisions. This study developed an artificial intelligence-driven business analytics framework for real-time fraud detection using the PaySim dataset containing 6,362,620 transactions, including 8,213 fraudulent cases. Nineteen pre-transaction features were generated, and four models—Random Forest, XGBoost, Support Vector Machine (SVM), and 1D Convolutional Neural Network (CNN)—were evaluated using a chronological 70% training, 15% validation, and 15% test framework. Random Forest achieved the best performance on the naturally imbalanced test set, obtaining a PR-AUC of 0.9999, precision of 99.87%, recall of 99.35%, and F1-score of 99.61%. The proposed framework also converts fraud-risk scores into operational decisions for transaction approval, manual review, or blocking. The findings demonstrate the potential of AI-driven business analytics for timely fraud-risk identification, although validation with real institutional transaction data is required before practical deployment.