Schedule of CPD Program : Vizualization and Interpretation of Machine Learning Models (2 CPD)

• Global interpretation with variable importance measures and partial dependence plots.
• Local analysis with local interpretable model agnostic explanations (LIME).
• The SHAP model for local interpretation.
• The LIME approach is illustrated with the R library "lime".

Abstract:

Machine learning (ML) models are often considered as “black boxes” due to their complex inner-workings. More advanced ML models such as random forests, gradient boosting machines (GBM), artificial neural networks (ANN), among others are typically more accurate for predicting nonlinear, faint, or rare phenomena. Unfortunately, more accuracy often comes at the expense of interpretability. This training focuses on recent approaches to interpret machine learning models. We first review global interpretation methods including partial dependence plots (PDP), Individual Conditional Expectation (ICE), Accumulated Local Effects (ALE), permutation feature importance, leave-one-covariate out (LOCO), H statistic and global surrogate models.  The second part of the module explores local approaches as the local interpretable model-agnostic explanations (LIME) and Shapley values (SHAP). The course is illustrated with R (package IML) and concepts are applied to interpret a neural network fitted to a credit default data set (package KERAS).  The R code shall be provided participants.