Decentralized tumor classification is a critical task in healthcare, especially for preserving patient privacy while enabling collaborative model training across multiple institutions. This paper presents a novel federated learning-based approach for classifying breast tumors as malignant or benign using the Breast Cancer Wisconsin (Diagnostic) dataset. The proposed model leverages distributed data from simulated clients, such as hospitals or diagnostic centers, and ensures data privacy by aggregating locally trained models without sharing sensitive data. Our experiments demonstrate that the federated learning model achieves superior performance compared to a centralized baseline, with an accuracy of 97.3%, precision of 96.1%, recall of 95.4%, and an F1-score of 95.8%. The model’s robustness is validated across varying numbers of clients and communication rounds, achieving stable convergence and high performance with as many as 50 clients. Additionally, the model shows resilience to class imbalance, maintaining an accuracy of 96.2% even with a benign-to-malignant ratio of 5:1. This study highlights the potential of federated learning for privacy-preserving medical applications, providing a scalable and effective solution for collaborative healthcare AI systems. The proposed approach ensures high diagnostic accuracy while safeguarding patient data, paving the way for real-world deployment in decentralized healthcare environments.