Interpretable Machine Learning with R
summary
Compares model-agnostic versus model-specific approaches using R packages and concrete examples: DALEX+vip for permutation-based importance, randomForest’s Gini issues, partial dependence and ALE (accumulated local effects) on the Boston data to show interactions, and LIME and SHAP for local explanations. Shows a medical case where a high-performance model used asthma as a proxy, warns about collinearity, notes breakDown is slow while iBreakDown faster gives similar instance attributions, and points to rule‑ensemble alternatives that build interpretability into training.
AI-written summary · about
chapters (22)
- 0:00Why "Interpretable" Machine Learning?
- 0:14In the beginning...
- 0:46Darkness falls...
- 1:30What constitutes the best medical treatment for recovery?
- 2:21What? - Data (Features/Instances)
- 2:49How well do you understand the data?
- 3:24Kind of Model Interpretation?
- 3:38Exploit Internal Logic
- 4:08Scope of Model Interpretation?
- 4:41What do you want to understand about the features and the model?
- 5:10Variable Importance Plot
- 7:47What about Interactions?
- 8:41Partial Dependence Plot
- 9:54What If?
- 10:05Ceteris Paribus Plot
- 10:38Local Interpretable Model-Agnostic Explanations
- 11:55Shapley Values (Coalition Attribution)
- 12:33Break Down (Feature Contributions)
- 13:23Foundations
- 13:52"Variable Importances"... Linearity Assumptions... Collinearity Computation Time
- 14:21Conclusion
- 14:51Rule Ensembles
related talks
34mInterpreting Machine Learning Models: Why and How!
44mInterpretable Machine Learning with Python Examples
01h 33mUnderstanding and diagnosing your machine-learning models
28mExplaining Machine Learning Models with LLMs
01h 36mOpen the Black Box: an Introduction to Model Interpretability with LIME and SHAP
46mExplaining a Machine Learning blackbox
26mStop Explaining Black Box Models, Use Interpretable Models Instead
01h 24mInpsect and try to interpret your scikit-learn machine-learning models