A Proposed Solution to Build a Breast Cancer Detection Model on Confidential Patient Data using Federated Learning

Due to the increasing number of privacy breaches of personal data there is a need for the development of methods that function along with the intent of preserving user privacy. Keeping this in mind we have proposed an algorithm using a federated approach to predict whether or not a patient is suffering from breast cancer, using data from multiple hospitals. This approach ensures that the user data is protected. The federated approach provides the hospitals with a safe and secure way to train their models without having to send their data to a central server. We have compared our approach with the standard approach to evaluate the performance of the federated approach. We determined that the federated learning model was able to achieve an accuracy comparable with the conventional model. There are advantages as well as limitations of our approach, that have been discussed further. This paper discusses an overall idea of federated learning, the past works done in this field, and our approach to implement a solution.