Comparative study of Federated learning models in a cross-silo scenario with Flower framework

Federated learning in a cross-silo scenario's encounters obstacles such as uneven performance across many medical fields, class imbalance, and data privacy concerns. These problems are difficult for traditional centralized machine learning techniques to solve while preserving the efficacy of the model and the confidentiality of the data. This study aims to examine various local training strategies in a privacy-preserving framework, address class imbalance issues, and evaluate the performance of federated learning models in several healthcare data sets. The flower framework is used to apply federated learning in three healthcare datasets: heart disease, breast cancer, and diabetes. Techniques such as SMOTE, ADASYN, and upsampling are used to reduce class imbalance. Both logistic regression and neural networks were used for local training. Various communication rounds were used to assess model performance using accuracy, precision, recall, F1 score, and loss measures. Across all datasets, the federated learning models performed better with more communication rounds. After 200 rounds, the breast cancer model had the best results, scoring 95% accuracy, 96% precision, 96% recall, and 96% F1 score. Accuracy rates for the diabetes and heart disease models were 80% and 81%, respectively. In local training, neural networks typically performed better than logistic regression. This study demonstrates the performance of federated learning in healthcare applications across several domains.