Beyond the Black Box: Interpreting ML Models with SHAP

Materials from PyData Berlin 2025 talk and Scipy 2025 poster on model interpretability using SHAP (SHapley Additive exPlanations).

Overview

This repository contains code and slides demonstrating how to use SHAP for explaining machine learning models across different domains:

Case Study 1: XGBoost classifier on tabular banking data Case Study 2: 1D CNN on time series sensor data for human activity recognition

Quick Start

1. Clone the repo

git clone https://github.com/ab93/explainable-ml-shap.git
cd explainable-ml-shap

2. Install dependencies using Poetry

poetry install -v

3. Open the notebooks

jupyter notebook

Case Studies

Bank Marketing (XGBoost + TreeSHAP)

• Dataset: Portuguese bank direct marketing campaigns
• Task: Predict term deposit subscription (binary classification)
• Model: XGBoost Classifier (AUC: 0.79)
• Explanation: TreeSHAP for fast, exact explanations

Human Activity Recognition (CNN + DeepSHAP)

• Dataset: Smartphone sensor data (accelerometer + gyroscope)
• Task: Classify 6 human activities (multiclass classification)
• Model: 1D CNN with PyTorch
• Explanation: DeepSHAP for neural network interpretability

Citing

If you want to cite my work, please cite the paper:

@article{basu2025,
  author = {Basu, Avik},
  title = {Explaining ML predictions with SHAP},
  journal = {Python in Science Conference, 2025},
  year = {2025},
  doi = {10.25080/mhum9729},
  url = {https://doi.org/10.25080/mhum9729}
}