Transparency and Privacy the Role of Explainable AI and Federated Learning in Financial Fraud Detection

Abstract

Originally, in the domain of financial fraud detection, relevance and applicability of Explainable AI (XAI) and Federated Learning (FL) provides a novel concept to mitigate the issues of opaque models and privacy breaches. This project uses the geographic data set from Kaggle called Paysim1 for testing the authenticity of the transaction focusing more on fraudulent activities. The existing system mainly incorporates DNN, RNN, and SGD but did not place much emphasis on Trust-aware Machine Learning. These methods are informative, nevertheless, they usually are not interpretable and it is necessary to save data in a centralized manner, which leads to privacy issues and model interpretability. On the other hand, the proposed system includes Decision Trees, Random Forest and Gradient Boosting Machines. The algorithms included are chosen because of their stability, and ease of interpretability and accuracy in data of high dimensionality. Moreover, Federated Learning is employed to realize privacy since training occurs on multiple decentralized devices but not the raw data. This approach ensures the privacy of the data whilst at the same time providing an environment for learning that enhances the training of the fraud detection models. Explainable AI and Federated Learning are proposed to address the main challenge of developing transparent and privacy- preserving solutions in the context of financial fraud detection.