Federated Learning has become a viable solution to train machine learning models on decentralized data without sharing sensitive information. Meanwhile, Blockchain ensures data integrity and transparency. In this work, we utilize the best of both worlds by implementing a distributed blockchain ledger within a federated learning system, named Fed-DL. Due to aggregation of non-IID data within federated setup, the model favors over-represented groups within clients. This introduces inequality in model’s predictive performance across different demographics, resulting in bias within the system. The proposed federated learning framework based on smart contract filters irregular or false updates from clients that negatively impact predictive analysis within demographic groups. This framework is also effective in reducing the server traffic by allowing limited updates which are validated through consensus for model aggregation at the server. Fed-DL not only helps in protecting sensitive client’s information but also fosters trust and equality among participants through transparency and record (updates) traceability. This approach enhances the overall quality of machine learning models while adhering to data protection laws and prompts fair and ethical practice. Results indicate that the proposed framework performs well in non-IID data scenarios and is also robust against poisoning attacks.