With the widespread application of decentralized data trading and federated learning, data privacy protection and system security have become pressing issues, particularly with the threat of poisoning attacks on model performance. This paper proposes a novel mechanism that combines decentralized validation and Shapley value-based contribution evaluation to defend against poisoning attacks and implement fair incentives. First, a decentralized validation and voting mechanism is designed, where multiple validator nodes evaluate model updates, ensuring that malicious updates do not affect global model training. Additionally, Shapley values are used to assess the contribution of each participant and fairly distribute rewards based on their contributions, thus incentivizing honest participation. Experimental results demonstrate that the proposed mechanism significantly enhances the system’s defense against poisoning attacks, while also improving computational efficiency and fairness. The results show that the integration of decentralized validation and Shapley values offers superior security and reduced unnecessary computation, providing an effective solution for decentralized data trading systems in federated learning applications.