Gestational weight gain prediction using privacy preserving federated learning.

Area of Science:

• Machine Learning
• Maternal Health
• Data Privacy

Background:

• Gestational weight gain prediction is vital for managing pregnancy risks.
• Early prediction allows for timely, manageable interventions.
• Centralized data storage for model training raises privacy concerns.

Purpose of the Study:

• To develop a privacy-preserving federated learning approach for early gestational weight gain prediction.
• To enable collaborative model training without sharing sensitive patient data.
• To assess the performance and privacy trade-offs of the federated approach.

Main Methods:

• Implemented a federated learning framework for collaborative model training.
• Ensured incremental model updates, avoiding eternal storage of local data.
• Evaluated the model's accuracy using mean absolute error.

Main Results:

• The proposed privacy-preserving federated model achieved a mean absolute error of 4.455 kgs.
• A centralized approach using individual data reached a mean absolute error of 2.572 kgs until day 140.
• The federated approach demonstrated effective privacy preservation with a minor performance reduction.

Conclusions:

• Federated learning offers a viable solution for privacy-preserving gestational weight gain prediction.
• Early prediction is achievable with acceptable performance using this method.
• This approach supports the development of machine learning algorithms for improved maternal care while safeguarding patient data.