A Scalable MLOps Framework for Intelligent Fraud Detection and Identification in Financial Transactions

The growing amount of digital financial transactions has presented considerable hurdles for accurate and flexible fraud detection. Machine learning (ML) approaches may detect complicated transaction patterns, but class imbalance, dynamic fraud behaviour, and effective model maintenance remain significant obstacles. The purpose of this study is to provide a scalable Machine Learning Operations (MLOps) platform that can intelligently detect and identify financial transaction fraud. The system includes data preprocessing, class balancing using the Synthetic Minority Over-sampling Technique (SMOTE), Random Forest (RF) feature selection, ML classification, model evaluation, explainable artificial intelligence (XAI), experiment tracking, and monitoring. The proposed technique was tested on a sample of 100,000 transactions from the Credit Card Transactions Fraud Detection Dataset. The four models developed and evaluated were XGBoost, HistGradientBoosting, LightGBM, and a Hybrid Stacking Model. LightGBM and HistGradientBoosting had the greatest recall of 82.50%, while the Hybrid Model had the highest accuracy of 99.49%, precision of 55.23%, and F1-score of 65.07%. The results demonstrate that an organised approach to model development, evaluation, and tracking may be achieved by integrating an MLOps framework with machine learning-based fraud detection. Additional research using diverse datasets and dynamic financial transaction environments may be built upon the proposed technique.