A Federated Learning based Approach for Heart Disease Prediction

Abstract

Cardiovascular diseases are the number one cause of death globally. Seventeen million deaths are recorded annually due to cardiovascular diseases. It is also estimated that four out of five cardiovascular deaths are due to heart failure. The existing machine learning methods to detect heart failure require data to be present in a centralized location. Due to data security and privacy, hospitals cannot share their patient’s data. Hence it is very difficult to store all data centrally. In this work, Federated learning is leveraged to train a shared model. Unlike the traditional machine learning approaches, the details regarding the patients are distributed in local databases. A shared model is computed by aggregating local updates from distributed clients. Logistic Regression and Support Vector Machine (SVM) algorithms are applied to train the model in client devices and only the updates are sent back to the centralized server. Experimental results have shown an accuracy of around 89% on the UCI benchmark dataset. The results of this experiment prove that federated learning could be applied where data privacy is a major concern. Federated Learning could be applied to process electronic health records (EHR) to build a shared model while keeping the data distributed across many geographic locations.

Ähnliche Arbeiten