Federated Learning to Preserve the Privacy of User Data

Abstract

In today’s world, the most pressing problem is the privacy of users’ data. It becomes even more critical when dealing with medical data, which is extremely sensitive. On the other hand, traditional machine learning (ML) algorithms require a single centralized source of data, which frequently compromises data privacy because data must be shared in order for the algorithm to be trained. As a result, federated learning is utilized to train ML algorithms on local private data sets spread across many sites. This also safeguards the privacy of the data. Our suggested method attempts to diagnose two urinary system diseases. We begin by training our model with logistic regression. Then, in order to simulate federated learning, we divided our data set into three sections. Then, each party trains the ML model on its local data set, and all of the updates from the local models are delivered to the trusted aggregator, which averages all of the updates. The averaged model is then sent to all individual parties at the start of each iteration. This entire system aids in the successful identification of two urinary system disorders by utilizing federated learning and distributed data principles in which data is not shared between individual parties and remains in distinct locations, considerably improving the privacy of such sensitive data.