Federated Learning

This is a repository for the Federated Learning project for the thesis "Federované učenie v zdravotníctve".
Working with Diabetes dataset. Which has 100 000 records and 9 attributes. After correlation analysis and feature selection I decided to use 6 attributes. The dataset is divided into 3 parts. 70% for training, 10% for validation and 20% for testing -> classic model part. For the federated part I used 2 clients. Server side has 10% of data for testing. Rest of the data is divided into 2 clients. Each client has 45% of data for training and 5% for validation. Due to imbalance of the dataset I used oversampling technique called KMeansSMOTE.

Classification report for classic model

	Precision	Recall	F1-Score	Support
No Diabetes	0.98	0.99	0.98	17991
Diabetes	0.84	0.74	0.79	1659
accuracy			0.97	19650
macro avg	0.91	0.86	0.88	19650
weighted avg	0.96	0.97	0.97	19650

Classification report for federated model (20 epochs, 5 rounds)

	Precision	Recall	F1-Score	Support
No Diabetes	0.98	0.99	0.98	18140
Diabetes	0.83	0.75	0.78	1660
accuracy			0.97	19800
macro avg	0.90	0.87	0.88	19800
weighted avg	0.96	0.97	0.96	19800

Technical University of Košice, Intelligent Systems. Department of Cybernetics and Artificial Intelligence 2024.