EXPLORE PUBLICATIONS BY COUNTRIES

Explainable AI (XAI), Deep Learning, Fraud Detection, Tokenization, SHAP.

The growth of electronic payments merely made it more imperative to uncover fraud. Tokenization, too, whereby security is based on replacing card information with temporary values, provides enough latitude for able fraudsters to exploit transactional trends. These are "black boxes," i.e., financial black boxes. This paper presents an open deep learning system to detect online credit card fraud in tokenized transaction systems like uniquely. We employed a feedforward deep neural network for classifying transactions as real or fraudulent. To tackle the challenge of explaining knowledge, we employed SHapley Additive exPlanations (SHAP) to explain features for each prediction in an interpretable manner. We trained and tested the model using a sample data set of 425 tokenized actual transactions. The data set includes tokenized card numbers, transaction amount, merchant names, and device IDs, etc. The whole system was implemented in Python with TensorFlow and Keras being used for neural network calculations and SHAP library being used for building explanations. The outcome shows not only that our model successfully identifies fraud but also a flawless depiction of its decision-making process and thus an optimal and reliable solution for banks.