Open the Black Box: an Introduction to Model Interpretability with LIME and SHAP
summary
Workshop demonstrates concrete claims and code: uses Eli5 for white‑box weights, LIME for local, model‑agnostic approximations, and SHAP (tree explainer) for global and local Shapley values. Example uses a Portuguese bank direct‑marketing dataset with class imbalance, shows ~83% accuracy but poor minority recall, surfaces a surprising proxy feature (number_of_employees) dominating predictions, notes LIME instability on images, and shows how to compute and visualize feature attributions and R² for trust.
AI-written summary · about
chapters (8)
- 0:00Workshop setup
- 3:35Model interpretability matters
- 12:53Eli5 library
- 35:57LIME for local explanations
- 59:59SHAP insights
- 1:15:30LIME for non-tabular data
- 1:23:11Key takeaways
- 1:25:04Q&A
related talks
31mBeyond the Black Box: Interpreting ML Models with SHAP
46mExplaining a Machine Learning blackbox
44mInterpretable Machine Learning with Python Examples
34mInterpreting Machine Learning Models: Why and How!
39mKeynote 3: How did you know? Explaining Black Box Model Predictions in Python
27mInterpretability and explainability in machine learning
35mTaking a peek under the hood - Interpreting black box model
16mBuilding Trust through Explainable AI