Restore full detailed data card and fix metadata

README.md

@@ -18,20 +18,74 @@ size_categories:
 
 ## Dataset Summary
 
-This dataset is derived from the Kaggle diabetic readmission dataset and is formatted for binary classification of 30-day hospital readmission.
+This dataset is derived from the Kaggle Diabetic Patients Readmission Prediction dataset and contains electronic health record (EHR) data from diabetic patient hospital encounters across the United States.
+
+It is formatted for binary classification of 30-day hospital readmission and is designed for evaluating predictive modeling and uncertainty quantification methods.
+
+## Source Data
+
+Original dataset:
+
+https://www.kaggle.com/datasets/saurabhtayal/diabetic-patients-readmission-prediction
 
 ## Prediction Task
 
-- Target: `readmit_30d`
-- `1`: readmitted within 30 days
-- `0`: otherwise
+Binary classification of 30-day hospital readmission.
+
+Target variable: `readmit_30d`
+
+- `1`: readmitted within 30 days (`<30`)
+- `0`: not readmitted within 30 days (`>30` or `NO`)
+
+## Dataset Files
+
+- `train.csv`: processed training split
+- `validation.csv`: processed validation split
+- `test.csv`: processed test split
+- `train_raw.csv`: raw training split
+- `validation_raw.csv`: raw validation split
+- `test_raw.csv`: raw test split
+- `cohort.csv`: identifiers + labels
+- `features.csv`: raw features
+
+## Dataset Statistics
+
+| Split | N | Negatives | Positives | Positive Rate |
+|---|---:|---:|---:|---:|
+| Train | 71236 | 63286 | 7950 | 0.1116 |
+| Validation | 15265 | 13561 | 1704 | 0.1116 |
+| Test | 15265 | 13562 | 1703 | 0.1116 |
+| Total | 101766 | 90409 | 11357 | 0.1116 |
+
+## Data Processing Pipeline
+
+- Replaced `"?"` with missing values
+- Created binary label `readmit_30d`
+- Stratified 70/15/15 split
+- Removed identifiers and target columns from features
+- Median imputation for numeric features
+- One-hot encoding for categorical features
+- Aligned feature columns across splits
+
+## Intended Use
+
+- Clinical prediction modeling  
+- Uncertainty quantification  
+- Risk-coverage analysis  
+- Decision-aware evaluation  
+
+## Limitations
+
+- Retrospective dataset  
+- Potential label noise  
+- Missing values  
+- Class imbalance  
+- Limited generalizability  
 
-## Files
+## Ethical Considerations
 
-- `train.csv`
-- `validation.csv`
-- `test.csv`
+This dataset is derived from de-identified clinical data and should not be used for clinical decision-making without validation.
 
-## Notes
+## Citation
 
-This dataset is intended for uncertainty quantification and clinical prediction research.
+Please cite the original Kaggle dataset when using this resource.