High-Precision Credit Card Fraud Detection on Imbalanced Data Using Random Forest and 1D Convolutional Neural Networks

Abstract

Credit card fraud has become a significant challenge for the financial industry, resulting in substantial monetary losses and eroding consumer trust. Detecting fraudulent transactions is particularly challenging due to the severe class imbalance and high dimensionality of transaction data. This study proposes a systematic pipeline for fraud detection, integrating stratified sampling, Synthetic Minority Over-sampling Technique (SMOTE), and comparative evaluation of Random Forest (RF) and 1D Convolutional Neural Network (CNN) models. The performance of both models is assessed using standard metrics, including Accuracy, Precision, Recall, F1-Score, and Area Under the Receiver Operating Characteristic Curve (AUC). Experimental results demonstrate that RF achieves high precision (99.45%) on unseen test data, ensuring reliable detection of legitimate transactions. In comparison, CNN achieves near-perfect recall (99.95%) on training data, indicating a strong capacity to identify fraudulent patterns. Temporal analysis of transaction data further reveals distinct patterns between legitimate and fraudulent activities, highlighting the predictive importance of the Time feature. The findings provide practical guidance for deploying machine learning models in real-world financial settings: RF offers a reliable solution for immediate implementation, whereas CNN presents a promising approach for future enhancement after further validation.