A Robust Diabetes Mellitus Prediction System Based on Federated Learning Strategies
Murat Kuzlu, Zhenxin Xiao, Maliha Tabassum, Ferhat Özgür Çatak · 2023
In recent years, diabetes mellitus has emerged as one of the most prevalent diseases worldwide. This disease can be caused by various factors, and individuals often visit diagnostic centers and consult with doctors to identify the root cause. Fortunately, Machine Learning (ML) techniques have proven effective in predicting and detecting diabetes. However, developing robust ML models requires access to large datasets obtained from various healthcare systems. This poses significant privacy and confidentiality concerns, as sharing personal health data can be risky. The primary objective of this study is to design a robust diabetes mellitus prediction system using an ML-based binary classification approach along with federated learning strategies. The aim is to develop a system that can deliver performance comparable to centralized learning while ensuring data privacy and confidentiality. The results demonstrate that federated learning can provide comparable performance to centralized learning, thus making it a promising candidate for developing robust diabetes prediction models. Additionally, the use of federated learning strategies can help overcome the privacy and confidentiality concerns associated with large healthcare datasets from various healthcare systems. The findings of this study suggest that machine learning, particularly federated learning, can significantly contribute to the development of effective diabetes prediction models while addressing privacy and confidentiality concerns.