SECURE E-COMMERCE TRANSACTION FRAUD DETECTION USING GRAPH NEURAL NETWORKS

In the modern digital economy, the surge in online financial transactions has led to a parallel increase in fraudulentactivities, posing a significant threat to individuals, businesses, and financial institutions. Traditional frauddetection systems often struggle with adaptability and accuracy, especially when faced with evolving fraudulentpatterns and highly imbalanced datasets. This paper presents a hybrid machine learning approach to detectfraudulent transactions by combining the strengths of the Random Forest Classifier and Graph Neural Networks(GNNs). The proposed system begins with a detailed preprocessing pipeline transforming raw transaction datainto structured form, handling missing values, encoding categorical data, and applying standardization. To addressthe class imbalance problem commonly present in fraud datasets, balancing techniques are applied to enhancemodel robustness. The Random Forest algorithm is used to identify key transaction features and perform initialfraud predictions. In parallel, a GNN-based model captures complex relationships between users and transactionelements, effectively uncovering hidden fraud networks that are typically missed by conventional methods. Themodel’s performance is evaluated using standard metrics, and visualizations are generated to illustrate featureimportance and fraud distribution. The results indicate a significant improvement in detection accuracy andinterpretability, making this approach highly effective for real-time fraud prevention in financial systems.