MCDM-Fed: an adaptive federated learning approach for enhanced distributed intelligence

As data-driven technologies continue to evolve, the need for secure and efficient machine learning models in distributed environments has become increasingly critical. Conventional machine learning approaches face significant challenges in such settings, including privacy concerns, non-IID data distributions, and class imbalance. This research introduces a novel model, MCDM-Fed (Multi-Criteria Decision-Making in Federated Learning), designed to address these issues by constructing a robust global model without centralizing data. MCDM-Fed identifies the best local client models based on a comprehensive set of performance metrics and their similarity to the ideal solution. An entropy-based divergence mechanism improves model aggregation by prioritizing models according to their performance. In addition, a data balancing mechanism is incorporated to alleviate the class imbalance at the client level. Extensive computational experiments on real-world datasets demonstrate the model’s effectiveness in addressing distributed data challenges. Unlike traditional machine learning models that degrade in federated environments, MCDM-Fed maintains high classification accuracy. Its robustness makes it a promising solution for diverse, privacy-sensitive applications.