AI-Driven Provenance Tracking for Mitigating Fraud in E-Commerce Platforms Using Neural Networks

This paper offers an overview of a novel AI-based provenance tracking system that can reduce fraud in e-commerce environments through its neural network. This system has been designed and its effectiveness has been validated on the four datasets – transaction records (500, 000 records), supply chain logs (300, 000 records), user behavioral patterns (1, 000, 000 records), and product images (100, 000 records). Preprocessing steps might include data cleaning, data normalization, and data transformation as we prepared and formatted the data for use in building machine learning models. CNNs were compared along with RNNs and models having a blend of CNN and RNN architectures. In this work, the hybrid model has been proved to give better performance with an accuracy of 96. % 2%, precision of 95. 8%, recall of 94. 88% on a precision level of 92%, a recall of 5%, and an F1 score of 95. This represents a 1% improvement over the previous results, and is indicative of the capabilities of the hybrid model to work with multiple data types for identification of fraudulent patterns. The system was implemented in a real-time e-commerce environment and had an average response time to the user of 150ms, the system was able to handle 200 TPS and the system up-time was 99. 8% It was able to identify 4500 fraudulent transactions out of 300 false and fraudulent positive transactions and 200 false and fraudulent negative transactions to a fraud prevention of 95 %. 5%. The research used ‘MATLAB’ for data pre-processing and model building, which allows for the development of more optimal and less resource-intensive models. The results prove that the system has a potential of adding a positive value in the ways of e-commerce security promotion and is a holistic approach to combating fraud.