Federated learning approaches for fuzzy cognitive maps to support clinical decision-making in dengue

Abstract

Federated learning is a distributed machine learning approach developed to guarantee the privacy and security of data stored on local devices. In healthcare, specifically in diseases of public health interest such as dengue, it is necessary to develop strategies that guarantee such data properties. Therefore, the aim of this work was to develop three federated learning approaches for fuzzy cognitive maps for the prediction of mortality and the prescription of treatment of severe dengue. The validation of the approaches was performed on severe dengue datasets from two dengue endemic regions in Colombia. According to the results, the use of federated learning significantly improves the performance of models developed in centralized environments. Additionally, the use of federated learning allows guaranteeing the privacy and security of each client’s data due to the local training of the models. Federated learning is a useful tool in healthcare because it guarantees the privacy and security of patient data. Our results demonstrated the ability of aggregated models to predict mortality and prescribe treatment for severe dengue.