A Comparative Study of Neural Network and Machine Learning on Privacy Preserving Federated Learning for Healthcare Applications

Abstract

The digital era in healthcare has revolutionized paper-based medical records into Electronic Health Records(EHRs). EHRs contain sensitive medical as well as identity information about the patient. Applying Machine Learning/ Deep Learning techniques to EHR can make the healthcare system smarter. However, the main issue is that preserving the patient’s privacy is paramount, hence, the Federated Learning(FL) approach is used to build a smart healthcare system and maintain the patient’s privacy. This study compares the Artificial Neural Network(ANN) and Logistic Regression(LR) in the FL environment. FL-ANN and FL-LR framework is designed with three hospitals and is executed independently with diabetes and CKD(Chronic Kidney Disease)datasets. The comparative study on these frameworks is done based on the performance metrics and the FL Process Time. The outcome of the study is that the FL-LR framework outperforms FL-ANN concerning FL-Process Time. The FL-LR with CKD dataset results in a better accuracy of 98.12% with FL Process Time of 10.5 seconds. For the diabetes dataset, FL-ANN resulted in an accuracy of 97.66% with a maximum FL Process Time of 846.9 seconds.

Ähnliche Arbeiten