Abstract:
With the increase in online financial transactions, the risk of fraud involving credit cards has grown many-fold. Most fraud detection systems that exist are designed to either stress speed or accuracy but hardly achieve both simultaneously, hence leaving financial institutions open to new and sophisticated fraudulent activities. This research, therefore, tries to fill this gap by developing a real-time fraud detection model that balances high-speed and accurate processing for timely and effective fraud prevention.
Fraudulent credit card transactions detection will be carried out using the Recurrent Convolutional Neural Network (RCNN) approach. The approach chooses the RCNN architecture for its ability to capture time-related patterns in transaction sequences. The feature extraction process in transactional data is supported by the use of convolutional layers. To clean and optimize the dataset the pre-processing steps include feature scaling and normalization along with outlier detection. Through supervised learning along with validation procedures and cross-validation techniques the model achieves stronger performance across multiple fraud patterns.
The RCNN model achieved 92% accuracy and obtained an AUC score of 0.91 while accurately differentiating fraudulent transactions from legitimate ones. Analysis of performance data shows the model's ability to manage uneven data distributions in credit card fraud datasets and its adaptability to different fraudulent patterns. The research findings confirm RCNN as an efficient system designed for detecting fraud operations in real-time. Optimization and refinement processes will lead to results that demonstrate enhanced reliability and scalability for real-world applications.
