A Federated Learning-Inspired Evolutionary Algorithm: Application to Glucose Prediction.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Computational Biology

Background:

• Federated Learning (FL) frameworks often struggle with data privacy and solution interpretability.
• Existing FL approaches may not efficiently address both privacy and interpretability simultaneously.
• The medical domain, particularly diabetes management, requires robust solutions for data privacy and interpretable predictions.

Purpose of the Study:

• To propose an innovative Federated Learning-inspired evolutionary framework.
• To directly employ an Evolutionary Algorithm (EA) for Federated Learning tasks.
• To simultaneously address data privacy and solution interpretability in Machine Learning.

Main Methods:

• A master/slave approach where slaves hold local data and use EAs to generate prediction models.
• The master aggregates locally learned models from slaves to form global models.
• Utilizing Grammatical Evolution for forecasting future glucose values in diabetic patients.

Main Results:

• The proposed framework demonstrates superior performance compared to approaches without local model exchange.
• Knowledge sharing through local models leads to improved generalization capabilities.
• Performance improvements include ~3.03% precision, 1.56% recall, 3.17% F1-score, and 1.56% accuracy.

Conclusions:

• The evolutionary framework effectively enhances data privacy and model interpretability in Federated Learning.
• The knowledge-sharing process is validated for developing personal diabetes management models.
• The proposed method shows statistically superior performance and generalization capability.