---
tags:
- healthcare
- clinical-ml
- diabetes
- readmission-prediction
- lightgbm
- lgbm
library_name: lightgbm
pipeline_tag: tabular-classification
---

# hospital-readmission-phase2-lgbm - Hospital Readmission Risk Prediction

## Model Description

This hospital-readmission-phase2-lgbm model predicts the risk of 30-day hospital readmission for diabetic patients. The model was trained on the UCI Diabetes 130-US Hospitals dataset with robust cross-validation and comprehensive evaluation.

**Task:** Hospital 30-Day Readmission Risk Prediction  
**Model Type:** Gradient Boosting Machine (LightGBM)  
**Training Date:** 2025-12-15 10:55:05  
**Environment:** kaggle (GPU)

## Performance Metrics

### Cross-Validation Results (5-Fold CV)

| Metric | Value |
|--------|-------|
| Mean ROC-AUC | 0.8399 ± 0.0055 |

### Final Test Set Results

#### Primary Metrics
| Metric | Value |
|--------|-------|
| ROC-AUC | 0.8424 |
| PR-AUC | 0.4000 |
| F1 Score | 0.1053 |

#### Classification Metrics
| Metric | Value |
|--------|-------|
| Precision | 0.6978 |
| Recall | 0.0569 |

#### Clinical Metrics
| Metric | Value |
|--------|-------|
| Sensitivity (TPR) | 0.0569 |
| Specificity (TNR) | 0.9969 |

## Model Visualizations

### ROC Curve
![ROC Curve](./roc_curve.png)

### Precision-Recall Curve
![Precision-Recall Curve](./precision_recall_curve.png)

### Confusion Matrix
![Confusion Matrix](./confusion_matrix.png)

### Calibration Curve
![Calibration Curve](./calibration_curve.png)

### Feature Importance
![Feature Importance](./feature_importance.png)

### Learning Curves
![Learning Curves](./learning_curves.png)

### Validation Curves
![Validation Curves](./validation_curves.png)

### Cross-Fold Metrics Comparison
![Metrics Comparison](./metrics_comparison_across_folds.png)

## Dataset Information

| Property | Value |
|----------|-------|
| Total Samples | 101,766 |
| Features | 113 |
| Development Set | 86,501 |
| Final Test Set | 15,265 |

## Training Configuration

### Evaluation Pipeline
- **Final Holdout Split:** Stratified split into development and test sets
- **Hyperparameter Search:** Grid search with 5-fold cross-validation
- **Nested Early Stopping:** Inner validation split within each fold
- **Final Evaluation:** Untouched holdout test set

### Best Hyperparameters

```python
{
  "n_estimators": 150,
  "learning_rate": 0.05,
  "num_leaves": 31,
  "max_depth": -1,
  "subsample": 0.7,
  "colsample_bytree": 0.7,
  "reg_alpha": 0.0,
  "reg_lambda": 0.1
}
```

## Training Details

- **Total Training Time:** 214.37 minutes
- **Hyperparameter Search Time:** 128.14 minutes
- **Cross-Validation Folds:** 5
- **Early Stopping:** Yes
- **Device:** GPU

## Usage

### Loading the Model

```python
import joblib
import pandas as pd

# Load the trained model
model = joblib.load('gradient_boosting_model.joblib')

# Load your preprocessed features
X_new = pd.read_csv('your_features.csv')

# Make predictions
predictions = model.predict(X_new)
probabilities = model.predict_proba(X_new)[:, 1]
```

### Feature Requirements

The model expects preprocessed features from the UCI Diabetes 130-US Hospitals dataset. Features include:
- Patient demographics (age, gender, race)
- Admission details (admission type, source, length of stay)
- Medical history (number of diagnoses, procedures)
- Medication information
- Lab results (A1c test results, glucose serum test)
- Previous utilization (outpatient, inpatient, emergency visits)

See `feature_importance.csv` for complete feature list and importance scores.

## Limitations and Biases

- **Domain-Specific:** Model is trained specifically for diabetic patient readmissions
- **Dataset Bias:** Training data from 130 US hospitals (1999-2008) may not generalize to all healthcare settings
- **Class Imbalance:** Dataset may have imbalanced readmission rates
- **Temporal Drift:** Healthcare practices have evolved since data collection
- **Geographic Limitation:** US-based dataset may not apply to other healthcare systems

## Ethical Considerations

This model is intended to assist healthcare providers in identifying patients at risk of readmission. It should:
- **NOT** be used as the sole basis for treatment decisions
- Be validated on your specific patient population before deployment
- Be monitored for fairness across different demographic groups
- Be regularly retrained with recent data to account for changing patterns

## Citation

```bibtex
@misc{hospital-readmission-phase2-lgbm,
  author = {Your Name},
  title = {LightGBM Model for Hospital Readmission Prediction},
  year = {2025},
  url = {https://huggingface.co/your-repo}
}
```

## Dataset Citation

```bibtex
@misc{strack2014impact,
  title={Impact of HbA1c Measurement on Hospital Readmission Rates: Analysis of 70,000 Clinical Database Patient Records},
  author={Strack, Beata and DeShazo, Jonathan P and Gennings, Chris and Olmo, Juan L and Ventura, Sebastian and Cios, Krzysztof J and Clore, John N},
  journal={BioMed Research International},
  volume={2014},
  year={2014},
  publisher={Hindawi}
}
```

## License

This model is released under the MIT License. The underlying dataset has its own license terms.

## Contact

For questions or issues, please open an issue in the repository.

---

**Disclaimer:** This model is for research and educational purposes. Always consult healthcare professionals for medical decisions.