FedEntropy: An Entropy-based Federated Random Forest Modeling Approach for Remote Diabetic Patient Monitoring Systems

Remote monitoring of diabetic patients emerges deeper into existence as COVID-19 fuels its persistent importance post-COVID era. Diabetic patients, in general, need continuous monitoring and prediction of glucose levels so that chronic conditions are improved by sending proactive messages to the concerned hospital authorities. To avoid data privacy issues while developing models, federated learning approaches were embodied in remote monitoring solutions in the past. However, the existing federated learning approaches are resource-inefficient. In this paper, an entropy-based federated random forest modeling approach named FedEntropy is designed. FedEntropy avoids compute nodes that do not have sufficient information to contribute to the collaborative learning process; it attempts to minimize the utilization of resources while enhancing data privacy and solving the prediction problem using Random Forest algorithms. Experiments were carried out at the IoT Cloud Research Laboratory to evaluate the efficiency of the FedEntropy approach with three learning models – Logistic Regressions, Support Vector Machines, and Random Forests. Results point out that the proposed FedEntropy based on the Random Forest algorithm minimized the memory utilization of computations to over 20000 MB.