Robust federated learning for cloud environments using evolutionary optimization and blockchain

The growing dependence on cloud-based distributed intelligence requires learning systems that can protect data privacy while maintaining efficiency. Federated learning has become an important method to train models without centralizing raw data, but its effectiveness is hindered by heterogeneous client distributions, adversarial risks, unstable convergence, and inefficient resource utilization. To overcome these challenges, this study introduces FedGenBlk, a federated learning framework that integrates genetic algorithm based optimization with blockchain-supported aggregation. The genetic algorithm dynamically tunes hyperparameters and guides client selection, ensuring stable convergence even under non-IID (non-independent and identically distributed) conditions. Meanwhile, blockchain offers tamper-proof and verifiable model aggregation, reducing the impact of malicious updates and Byzantine failures. Experiments conducted on the EMNIST dataset demonstrate that FedGenBlk achieves 91.2% accuracy, with only 1.9% variance between IID and non-IID settings, and sustains robustness under poisoning attacks with just a 3.1% accuracy drop. Additional validation on CIFAR-10 and Fashion-MNIST datasets confirms consistent performance advantages across different data modalities. These results highlight the framework's effectiveness in enhancing reliability, security, and adaptability of federated learning in cloud infrastructures, making it particularly suitable for healthcare, IoT, financial services, and cross-organizational collaborative intelligence applications.