Interpretable Machine Learning - Online Course

This seminar will give you the tools and skills to:

• • Understand why machine learning (ML) models often outperform traditional models and why interpretability is essential.
  • Differentiate between intrinsically interpretable models and black-box models.
  • Use global interpretation tools such as partial dependence plots (PDPs), ALE plots, feature importance, and surrogate models.
  • Use local interpretation tools such as individual conditional expectation (ICE) curves, LIME, anchors, and SHAP values.
  • Interpret direction, magnitude, and heterogeneity of effects using model-agnostic tools.
  • Apply bootstrapped uncertainty methods to assess statistical confidence in ML interpretation.

To participate in the hands-on exercises, you are encouraged to use a computer with the most recent version of R, RStudio, and relevant ML packages installed. The list of relevant packages will be distributed prior to the first course meeting. Equivalent Python code will be provided where applicable.

You should be comfortable performing basic tasks in R, such as importing and coding data, making simple plots, and estimating regression models.

If you’d like to take this course but are concerned that you don’t know enough R, there are excellent online resources for learning the basics. Here are our recommendations.

This seminar is designed for applied researchers across various fields, including medical research, public health, biostatistics, economics, government analysis, industry data science, and consulting, who utilize predictive models and require tools for explanation, transparency, and effective communication.

You should have a basic familiarity with statistical modeling (e.g., regression) and data analysis in R. No prior experience with machine learning is required.

Foundations of interpretable machine learning

• • Why interpretability matters: transparency, trust, fairness, and reproducibility
  • Overview of model complexity and accuracy–interpretability tradeoff
  • Intrinsically interpretable models: linear/logistic regression, decision trees, sparsity via LASSO
  • Introduction to “black-box” models: random forests, gradient boosting (XGBoost), neural networks
  • Global vs. local interpretability and model-agnostic frameworks

Post-hoc interpretation methods

• • Global interpretation tools:
    • Partial dependence plots (PDPs)
    • Accumulated local effects (ALE) plots
    • Feature interaction measures (e.g., H-statistic)
    • Permutation feature importance with uncertainty
  • Local interpretation tools:
    • Individual conditional expectation (ICE) curves
    • Local surrogate models (LIME)
    • Anchors and scoped rules
    • Counterfactual explanations

Applications, uncertainty, and advanced topics

• • SHAP and Shapley values: theory, intuition, and practice
  • SHAP-based feature importance and model explanation
  • Bootstrapped uncertainty for interpretability tools:
    • Confidence intervals for PDPs and ICE curves
    • Bootstrapped SHAP-based marginal effects
  • Evaluating model stability and robustness

The fee of $995 USD includes all course materials.

PayPal and all major credit cards are accepted.

Need funding from your employer or institution? Use our funding request template.

Our Tax ID number is 26-4576270.