| tags: | |
| - tabular-classification | |
| - healthcare | |
| - synthetic-data | |
| license: mit | |
| # HFrEF prediction model (synthetic EKG data) | |
| XGBoost classifier that predicts **Heart Failure with Reduced Ejection Fraction (HFrEF)** from features derived from synthetic 12-lead EKG data and a synthetic lab panel. | |
| **All training data is synthetic** - generated by our own generator. This model is a course/educational artifact and must NOT be used for any real clinical decision. | |
| ## Model version | |
| `model_20260925_130635.joblib` | |
| ## Evaluation metrics | |
| | Metric | Value | | |
| |---|---| | |
| | AUROC | 0.9349 | | |
| | AUPRC | 0.4572 | | |
| | Brier score | 0.0579 | | |
| | ECE (calibration) | 0.0534 | | |
| Confusion matrix (rows = true, cols = predicted): | |
| ``` | |
| [[342, 31], [13, 14]] | |
| ``` | |
| ## Important note on the label | |
| The label (HFrEF: ejection fraction < 40) is deliberately NOT a feature. Ground truth arrives on a delay in a separate outcomes table, so the model never sees the answer at prediction time. This is what makes the AUROC honest (high but not perfect) and the monitoring meaningful. | |
| ## How to load | |
| ```python | |
| from huggingface_hub import hf_hub_download | |
| import joblib | |
| path = hf_hub_download("anastasiyayudo/HFrEF", "model_20260925_130635.joblib") | |
| model = joblib.load(path) | |
| ``` | |