Table 5.
Federated machine learning implementations in diabetes prediction.
| Ref | Model | Data Used | Performance |
|---|---|---|---|
| [116] | FL deep neural network | SFU prototype swept-source OCTA RTVue XR Avanti (OptoVue, Inc.) Angioplex (Carl Zeiss Meditec) PLEX Elite 9000 (Carl Zeiss Meditec) | Performance is comparable to conventional DL models |
| [117] | Not identified | Health Facts EMR Data dataset from Cerner | Performance is similar to the gold standard of centralized learning |
| [118] | FL convolutional neural network (CNN) FL multilayer perceptron (MLP) | Generated by simulator | FL-CNN recall: 99.24% FL-MLP recall: 98.69% performed better than traditional DL |
| [119] | Not identified | Fine-Grained Annotated Diabetic Retinopathy (FGADR) dataset [120] | Accuracy: 72% |
| [121] | Not identified | Private data collected from different healthcare facilities | - |
| [122] | Standard FL FedAVG FedProx | EyePACS [123] Messidor [124] IDRID [125] APTOS [126] University of Auckland (UoA) [127] | Standard FL Accuracy: 92.19% FedAVG Accuracy: 90.07% FedProx Accuracy: 85.81% |