Hospital Readmission — Federated SVM (SGD) Report
Linear SVM via SGD with FedAvg weight averaging. Features standardized locally per client.
Privacy: Only model weights are shared — no raw patient data leaves each clinic.
Metrics
| Metric | Value |
|---|
| N Samples Train | 626 |
| N Samples Val | 157 |
| N Features | 132 |
| Federated Rounds | 10 |
| Train Accuracy | 0.6789 |
| Val Accuracy | 0.6369 |
| Train Auc | 0.7191 |
| Val Auc | 0.5639 |
| Val Precision | 0.1429 |
| Val Recall | 0.4706 |
| Val F1 | 0.2192 |
Confusion Matrix
| Predicted: Not Readmitted | Predicted: Readmitted (<30d) |
|---|
| Actual: Not Readmitted | TN=92 | FP=48 |
| Actual: Readmitted | FN=9 | TP=8 |
Classification Report
precision recall f1-score support
Not Readmitted 0.91 0.66 0.76 140
Readmitted <30d 0.14 0.47 0.22 17
accuracy 0.64 157
macro avg 0.53 0.56 0.49 157
weighted avg 0.83 0.64 0.70 157
Training History
ROC Curve