Federated Learning: A Novel Framework for Parkinson’s disease Detection

Abstract

Inadequate dopamine production from brain dopamine-producing cells leads to Parkinson's disease which causes dysfunction of body movements. The main disability factors of this disorder include impaired mobility together with impaired speech and vision difficulties while excretory function disruptions emerge as well. The detection algorithms for Parkinson's disease based on machine learning and deep learning show noteworthy progress yet they depend mainly on concentrated data collection systems that present major difficulties. The combination of data security risks and medical information exposure to hackers together with excessive data transfer expenses towards centralized repositories results in major patient privacy and data protection implications. The purpose of this work involves implementing federated learning techniques for Parkinson's disease diagnosis because traditional machine learning methods lack robust data security mechanisms and substantial privacy protections and centralized data computational limitations. The research uses different classification methods in a Federated Learning context to accurately identify Parkinson's disease through proper resolution of detected limitations. Model implementation and result analysis constitute the final parts of the methodology alongside data collection, preprocessing and exploration stages. The research utilized and evaluated Artificial Neural Network (ANN) together with Ridge Classifier and Logistic Regression. The Ridge Classifier executed as the most effective model by reaching 87.17% overall classification accuracy. The assessment shows that federated learning delivers superior performance when compared to conventional machine learning with minor benefits while overcoming issues of classic machine learning approaches. © 2026 IEEE.