A Secure and Scalable Peer-to-Peer Federated Learning Approach for Handling Veracity in Big Data

A massive source of terabytes of data is produced consistently from advanced data frameworks and innovation such as distributed/cloud computing, mobile network devices, and the Internet of Things. However, maintaining data veracity in handling such vast and diverse datasets presents significant challenges. In response to these challenges, Federated Learning (FL) emerges as a promising candidate to address the veracity issues in big data analytics. The distributed approach of FL not only ensures data privacy but also mitigates the risk of data breaches and unauthorized access, enhancing data security and veracity. In this paper, we present a Secure and Scalable Peer-to-Peer FL (P2P-FL), a novel approach that tackles the challenges of maintaining data veracity while simultaneously enhancing security and scalability through the utilization of multi-level encryption-decryption processes. This approach eliminates the need for a central server and allows federates to communicate directly, optimizing the efficiency of FL. Differential privacy techniques further ensure individual data privacy. Overall, our approach advances the effectiveness and security of machine learning models in privacy-sensitive environments. In our study, we conducted a comprehensive analysis traditional FL and Peer-to-Peer P2P-FL. We evaluated the performance based on accuracy, loss, and F1 score metrics. The results showcased the effectiveness of our Secure and Scalable Peer-to-Peer FL approach.