AI FOR REAL-TIME FRAUD DETECTION IN FINANCIAL TRANSACTIONS: LEVERAGING MACHINE LEARNING ALGORITHMS FOR EARLY DETECTION OF ANOMALIES

Abstract

This study presents a machine learning-based framework for real-time fraud detection in financial transactions, aiming to enhance the security and efficiency of financial institutions. The proposed framework integrates supervised and unsupervised machine learning models, including decision trees, support vector machines (SVM), k-means clustering, and autoencoders, to identify fraudulent activities by detecting anomalies in transaction data. The results show that the SVM model achieved the highest detection accuracy (92.1%) with a low false positive rate (3.5%), demonstrating its effectiveness in detecting known fraud patterns. In addition, unsupervised techniques, such as autoencoders and k-means clustering, enabled the detection of previously unseen fraud patterns, further improving the framework's ability to adapt to new fraud tactics. The framework performed real-time fraud detection with mitigating times of 95.6 ms while maintaining stable throughput during increasing transaction volumes which exceeded traditional fraud detection methods.  The use of ensemble techniques specifically XGBoost helped improve model accuracy and maintained low false positive rates.  The research emphasizes the requirement to optimize machine learning models that should produce fewer false positives to preserve valid transactions.  Applications included in the framework consist of interpretability tools that enhance openness and trust in decision-making procedures.  Machine learning demonstrates its power to track the dynamic nature of financial fraud through an effective system which detects fraud in real time.