Privacy-Enhanced Federated Learning for Rare Genetic Disorder Classification with EHR

Utilizing machine learning and deep learning for disease classification has gained widespread popularity, yet it encounters significant challenges. Rare genetic disorders pose a data scarcity issue, as individual hospitals often lack sufficient data for accurate modelling. Collaborative data aggregation for such rare diseases raises data privacy concerns. Additionally, complex domains like genetic disorders require multifaceted data, including family history, medication history, and medical records. To address these challenges, this research explores federated learning on electronic health records (EHR) for the classification of rare genetic disorders. The study employs a dataset with a diverse range of patient information and leverages neural networks in a federated learning setup. The results demonstrate a 6% improvement in accuracy compared to individual hospital models, highlighting the effectiveness of this approach. Furthermore, the study provides insights into the trade-off between accuracy and privacy, emphasizing the importance of federated learning in safeguarding patient data while enhancing disease classification accuracy in the context of rare genetic disorders.