Multi-Entity Real-Time Fraud Detection System using Machine Learning: Improving Fraud Detection Efficiency using FROST-Enhanced Oversampling

Abstract

Fraudulent transactions pose a significant threat to financial institutions and e-commerce platforms. Machine learning models, trained on historical labeled data (fraudulent vs. legitimate transactions), are often employed to identify and prevent fraud. However, real-world datasets frequently exhibit class imbalance, where fraudulent transactions (minority class) are significantly outnumbered by legitimate transactions (majority class). Machine learning models may perform poorly as a result of this imbalance, underestimating fraud and favouring the majority class. This paper proposes a novel approach to address class imbalance and improve fraud detection accuracy. We explore the implementation of FROST (Feature space RObust Synthetic saTuration) oversampling, a technique specifically designed to generate synthetic samples for the minority class. The FROST function leverages the k-nearest neighbors (KNN) algorithm and a user-defined amplification factor (m) to create synthetic data points that closely resemble existing minority class instances. We integrate the FROST-enhanced oversampling technique into the machine learning pipeline for fraud detection. The paper evaluates the effectiveness of this approach compared to traditional oversampling methods and analyzes its impact on classification accuracy metrics.