A decision-theoretic framework for building allocation-aware healthcare AI systems that optimise for fairness and resource efficiency alongside predictive accuracy.
This framework implements the research presented in "Beyond Accuracy: A Decision-Theoretic Framework for Allocation-Aware Healthcare AI" (Ferzana, 2026), published in the Artificial Intelligence and Intelligent Healthcare proceedings at Imperial College London.
- Allocation-aware loss functions that explicitly model resource constraints and fairness objectives
- Decision-theoretic evaluation metrics beyond accuracy
- Fairness-utility trade-off analysis for healthcare resource allocation
- Practical implementations for common healthcare allocation scenarios
# Clone the repository
git clone https://github.com/Rifa-111/allocation-decision-framework.git
cd allocation-decision-framework
# Install in development mode
pip install -e ".[dev]"
# Or install from PyPI (once published)
pip install allocation-decision-frameworkfrom allocation_framework import AllocationModel, FairnessLoss
# Initialise model with fairness constraints
model = AllocationModel(
fairness_weight=0.3,
resource_constraint=100, # e.g., 100 treatment slots
fairness_metric='allocation_parity'
)
# Train with allocation-aware objectives
model.fit(X_train, y_train, allocation_targets=targets)
# Evaluate beyond accuracy
results = model.evaluate(X_test, y_test)
print(results.fairness_score)
print(results.allocation_efficiency)
print(results.predictive_accuracy)Explore jupyter notebooks in examples/:
01_introduction.ipynb— Core concepts and fairness-accuracy trade-offs02_synthetic_allocation.ipynb— End-to-end example with synthetic data03_fairness_metrics.ipynb— Computing and interpreting fairness metrics04_resource_constraints.ipynb— Modeling real-world resource limits05_comparing_approaches.ipynb— Allocation vs. accuracy-only baselines
If you use this framework, please cite the original paper:
@inproceedings{ferzana2026beyondaccuracy,
author = {Ferzana, Rifa},
title = {Beyond Accuracy: A Decision-Theoretic Framework for Allocation-Aware Healthcare AI},
booktitle = {Artificial Intelligence and Intelligent Healthcare},
year = {2026},
series = {Lecture Notes in Computer Science},
volume = {16877},
pages = {XXX--XXX},
publisher = {Springer},
url = {https://link.springer.com/chapter/10.1007/978-3-032-35393-1_24}
}Healthcare AI systems must balance predictive accuracy with fair resource allocation. We present a decision-theoretic framework that explicitly models allocation constraints and fairness objectives, moving beyond single-metric evaluation. Our approach quantifies fairness-accuracy trade-offs and provides principled methods for practitioners to navigate allocation decisions in resource-constrained settings.
✅ Allocation-aware optimisation — Loss functions that encode resource and fairness constraints
✅ Flexible fairness metrics — Multiple fairness definitions (parity, equality of opportunity, etc.)
✅ Decision support tools — Trade-off visualization and sensitivity analysis
✅ Production-ready — Type hints, logging, and error handling
✅ Reproducible — Full examples and benchmark datasets included
allocation-decision-framework/
├── allocation_framework/
│ ├── __init__.py
│ ├── models/ # Core allocation models
│ ├── metrics/ # Fairness & allocation metrics
│ ├── losses/ # Custom loss functions
│ ├── visualization/ # Trade-off plots & analysis
│ └── utils/ # Utilities & helpers
├── examples/ # Jupyter notebooks
├── tests/ # Unit & integration tests
├── docs/ # Full documentation
└── setup.py
- Python 3.8+
- NumPy, Pandas, Scikit-learn
- PyTorch (optional, for deep learning variants)
- Jupyter (for tutorials)
See requirements.txt for full dependencies.
We welcome contributions! See CONTRIBUTING.md for guidelines.
pip install -e ".[dev]"
pytest tests/
black allocation_framework/MIT License — see LICENSE file for details.
- Author: Rifa Ferzana
- GitHub: @Rifa-111
- Issues: GitHub Issues
Published: August 2026 at AIiH 2026, Imperial College London
Paper DOI: 10.1007/978-3-032-35393-1_24