Privacy Preservation Using Federated Learning for Credit Card Transactions

Data leakage has become one of the crucial problems in the modern day world,in which the personal details of the customers including their credit card details.healthcare information, purchase history,geo-location data etc.,are stolen by hackers from the data furnishers who store and process these data. An efficient Credit card scoring system is essential for the benefits of both the bank and card-holders. Furthermore the banks must not share their data due to security reasons leading it difficult for data scientists to get any insights.It is also found that the data from devices are very skewed and not uniformly distributed. This might cause the model to bias towards one of the classes. Inorder to deal with this issue the data is partitioned with a fixed skew and then processed in the runners. Different from the traditional learning techniques where we assume data is identically distributed across all devices and centrally process them.federated learning trains the model in a decentralized way in local machines so the data remains safe.In this paper the federated Averaging Algorithm is implemented on a real-time credit card transaction dataset and the results shows that FedAvg outperforms the local models in both the case of uniform/iid and non-iid datasets. Keywords: Credit card fraud. Non-iid dataset. Federated learning - Federated Averaging Algorithm