Resolve accuracy drops in federated learning global aggregation caused by non-IID data and client drift using weighted averaging, FedProx regularization, or adaptive learning rates.