GPE (Greedy-Prune-Explain) is a novel method for generating minimal, interpretable local explanations for decision tree predictions. Unlike existing methods like LIME, SHAP, or Anchors, GPE leverages the inherent structure of decision trees to produce explanations that are:
- ✅ Minimal — Contains only essential conditions (1-2 instead of 5)
- ✅ Precise — 99.4% precision on real-world financial data
- ✅ Fast — 48x faster than LIME, 19x faster than Anchors
- ✅ Actionable — Simple IF-THEN rules like "income < 50000 AND debt_ratio > 0.4"
Tested on 3 financial datasets (632K records total):
| Method | Time (ms) | Complexity | Precision | Speedup |
|---|---|---|---|---|
| GPE-Core | 4.4 | 1.4 | 99.4% | 48x |
| GPE-IT | 3.0 | 1.4 | 97.9% | 71x |
| LIME | 213 | 5.0 | — | 1x |
| Anchors | 82 | 0.7 | 99.2% | 3x |
All results are statistically significant (p < 0.001)
pip install gpe-frameworkOr install from source:
git clone https://github.com/vladdehtiarov/gpe-framework.git
cd gpe-framework
pip install -e .from sklearn.tree import DecisionTreeClassifier
from sklearn.datasets import load_iris
from gpe import GPEExplainer
# Load data and train model
iris = load_iris()
X, y = iris.data, iris.target
model = DecisionTreeClassifier(max_depth=5)
model.fit(X, y)
# Create explainer
explainer = GPEExplainer(
model=model,
feature_names=iris.feature_names,
X_train=X,
min_precision=0.95
)
# Explain a prediction
explanation = explainer.explain(X[0])
print(explanation)Output:
============================================================
GPE Explanation (method: GPE)
============================================================
Prediction: 0
Rule: petal length (cm) <= 2.45
------------------------------------------------------------
Precision: 100.00%
Coverage: 33.33%
Complexity: 1 conditions
Reduction: 75.0% (4 → 1 conditions)
============================================================
print(explanation.to_natural_language())The model predicts 'setosa' because petal length is at most 2.45.
This explanation covers 33.3% of similar cases with 100.0% accuracy.
from gpe import (
GPEExplainer, # Standard (fast, greedy)
GPEInformationTheoretic, # Uses mutual information for pruning
GPECounterfactual, # Adds counterfactual explanations
GPEOptimal, # Exhaustive search for minimal rule
GPEEnsemble # For Random Forest, XGBoost
)
# GPE-IT: Uses mutual information I(condition; prediction)
gpe_it = GPEInformationTheoretic(model, feature_names=features, X_train=X)
explanation = gpe_it.explain(x)
# GPE-CF: Includes counterfactual explanation
gpe_cf = GPECounterfactual(model, feature_names=features, X_train=X)
cf_explanation = gpe_cf.explain_with_counterfactual(x)
print(f"To change the decision: {cf_explanation.changes}")GPE operates in three phases:
Extract the full decision path from root to leaf:
Root → income <= 50000 → debt_ratio > 0.4 → ... → Leaf (denied)
Iteratively remove conditions that don't affect precision:
while conditions > 1:
for condition in rule:
precision_without = calculate_precision(rule - condition)
if precision_without >= threshold:
remove(condition)Return the minimal rule with metrics:
- Precision — Accuracy for instances satisfying the rule
- Coverage — Proportion of dataset satisfying the rule
- Complexity — Number of conditions
from gpe import (
precision_score,
coverage_score,
complexity_score,
fidelity_score,
stability_score
)
# Evaluate explanation quality
precision = precision_score(explanation, model, X)
coverage = coverage_score(explanation, X)
complexity = complexity_score(explanation)- GPE-Core — First local explanation method specifically designed for decision trees
- GPE-IT — Novel use of mutual information I(condition; prediction) for condition selection
- Theoretical guarantees — Proven precision bounds and O(n·d) complexity
- Practical efficiency — 48x faster than LIME on real data
gpe-framework/
├── gpe/
│ ├── __init__.py # Public API
│ ├── core.py # GPEExplainer
│ ├── novel_methods.py # GPE-IT, GPE-CF (scientific contribution)
│ ├── variants.py # GPEOptimal, GPEWeighted
│ ├── explanation.py # Data structures
│ ├── metrics.py # Evaluation metrics
│ ├── tree_utils.py # Tree utilities
│ └── visualization.py # Plotting functions
├── tests/ # Unit tests
├── experiments/ # Benchmark scripts
└── docs/ # Documentation
Contributions are welcome! Please read our contributing guidelines and submit pull requests.
# Install development dependencies
pip install -e ".[dev]"
# Run tests
pytest tests/ -v
# Format code
black gpe/MIT License - see LICENSE file for details.
Vladyslav Dehtiarov
- Email: vvdehtiarov@gmail.com
- ORCID: 0000-0002-1578-8588
- Affiliation: Sumy State University, Ukraine
If you use GPE in your research, please cite:
@article{dehtiarov2025gpe,
title={Greedy-Prune-Explain: Minimal Local Explanations for Decision Tree Predictions},
author={Dehtiarov, Vladyslav and Borovyk, Valentyna},
journal={International Journal of Artificial Intelligence Research},
year={2025}
}