Explainable AI (XAI) is a vital subfield of artificial intelligence that aims to make the inner workings of machine learning models understandable. This repository offers a beginner-friendly yet insightful introduction to SHAP (SHapley Additive exPlanations), one of the most powerful tools in the XAI toolbox.
We demonstrate how SHAP can be applied to interpret model predictions on an income prediction problem using Python, scikit-learn, and shap.
# Clone the repository
git clone https://github.com/Harrypatria/ExplainableAI.git
cd ExplainableAI
# (Optional) Create and activate a virtual environment
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
# Install dependencies
pip install -r requirements.txt
# Run the Jupyter notebook
jupyter notebook- 🔍 Intuitive XAI Breakdown: Covers model-specific, model-agnostic, and hybrid interpretability approaches.
- 📈 Real Dataset: Demonstrates SHAP on an income prediction model with real-world features.
- 🧮 SHAP Visualisations: Gain insights into local and global feature importance.
- 🧠 Model Transparency: Learn how to build trust with stakeholders using interpretable ML.
- 📚 Educational Focus: Suitable for both industry professionals and students.
- Introduction
- XAI Terminology and Classification
- SHAP Overview
- Income Prediction Example
- Installation and Dependencies
- Usage Instructions
- License
- Acknowledgements
As machine learning becomes more pervasive, understanding its predictions is critical. XAI allows data scientists to ensure transparency, fairness, and accountability. SHAP is especially useful for quantifying the contribution of each feature to individual predictions.
- Model-Specific Methods: Tailored to specific algorithms (e.g., decision trees).
- Model-Agnostic Methods: Work with any model post-training (e.g., SHAP, LIME).
- Hybrid Methods: Combine strengths of both approaches.
SHAP applies game theory (Shapley values) to explain predictions. It ensures consistency and local accuracy, making it a gold standard in interpretability. You’ll learn:
- Feature importance
- SHAP summary plots
- Force plots for individual predictions
Given features like age, education, and occupation, we train a model to predict individual income. SHAP is then used to:
- Visualise global and local feature contributions
- Explore individual predictions
- Understand model behaviour
- Python 3.8+
scikit-learnshappandas,numpymatplotlib,seaborn,jupyter
This project is licensed under the MIT License. See the LICENSE file for details.
- SHAP by Scott Lundberg et al.
- UCI Adult Income Dataset
- Open-source contributors in XAI research
