MOFO-FL: A Multi-Objective Federated Optimization Framework for Privacy-Preserving Healthcare Cloud Cost Management

Abstract

The integration of cloud computing in modern healthcare information systems has greatly contributed to data scalability, real-time analytics, telemedicine delivery, and efficient management of EHRs. Nonetheless, the dynamic nature of the workload in healthcare often causes inefficiencies in cloud resource management and increased costs for cloud operations. Also, centralized machine learning models employed in the process of optimizing cloud resources have been identified to pose major threats when it comes to the privacy and security of patients' health data. With that in mind, this paper introduces MOFO-FL which stands for Multi-Objective Federated Optimization Framework for Privacy-Preserving Healthcare Cloud Cost Management. The developed system is based on combining FL with MOFO to allow for collaborative prediction of cloud costs while at the same time optimizing cloud resources in multiple healthcare centers without disclosing patients' sensitive health data. This is achieved by employing federated deep learning with FedAvg aggregation strategy for predicting cloud cost. Also, a multi-objective fitness function is used to optimize cloud cost, cloud execution time, CPU usage, and storage capacity. Evaluation via experiments proves fast convergence of the federated model with very low levels of prediction errors and improved cloud resource usage. XAI-based analysis reveals that the CPU needs and execution time are two parameters which have the highest impact on cost estimation in clouds. Comparative study conducted in comparison with existing techniques like genetic algorithm, particle swarm optimization, ant colony optimization, and whale optimization algorithm highlights the fact that the presented approach offers significantly better performance with the cloud cost savings reaching up to 99.96%.