Federated Knowledge Transfer and Interpretable Decision-Tree Analytics for IoMT Healthcare
TL;DR
The proposed federated knowledge transfer and interpretable decision-tree analytics for enabling privacy-preserving clinical prediction makes use of locally learned decision trees to generate clinical decision-path features and federated knowledge transfer on the level of features.
Abstract
As Internet of Medical Things (IoMT) enables continuous healthcare monitoring, it has also created major issues with respect to data privacy, distributed learning, and model interpretability in healthcare organizations. To address these issues, the proposed federated knowledge transfer and interpretable decision-tree analytics for enabling privacy-preserving clinical prediction. This approach makes use of locally learned decision trees to generate clinical decision-path features and federated knowledge transfer on the level of features. The fusion of global-local features makes it even better to make the predictions more reliable by retaining client-specific information. The proposed framework is experimented with MIMIC-III clinical dataset in a distributed IoMT environment. Experimental results show that the accuracy rate is 92.3%, precision rate is 91.8%, recall rate is 92.6%, F1-score is 92.2%, and AUC-ROC value is 0.963.