Comparative Study of Federated Learning Vs Centralized Learning

It is widely accepted that when data privacy is a concern, Federated Learning is preferable than the traditional approach of centralized learning. Deciding between the two approaches necessitates considering computational resources, scalability, and privacy concerns. The comparative study demonstrates that federated learning outperforms centralized learning in accuracy, precision, and recall, effectively handling distributed data and privacy preservation. Further research is recommended to identify specific scenarios where each method excels and understand performance disparities. Federated learning is ideal for privacy-sensitive applications, benefiting from its privacy-preserving advantages and decentralized training, while centralized learning remains suitable for scenarios with centralized data and lower privacy concerns due to its simplicity and effectiveness. The study is conducted using TensorFlow and keras. Federated learning has more performance level when compared to centralized learning, which has a comparison proportion of 85% and 50%, respectively