Taking a peek under the hood - Interpreting black box model
summary
A model-agnostic explanation method SHAP computes per-feature contributions by estimating Shapley values and claims three properties: missingness, local accuracy and consistency. Applied to an XGBoost churn/redemption model it identifies fund fees as the top driver, produces per-instance explanations, clusters similar SHAP vectors to reveal groups (e.g. medium-net-worth clients reacting to bond-index moves), and shows interactions like tenure × fee level. It works but is computationally intensive; sampling the data is the practical workaround.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 2:25Short story
- 3:54Why interpretation is important
- 6:45Simple model
- 16:34Whats next
- 16:47Background
- 17:57Results
- 20:41Clustering
- 21:52Interactions
- 23:19What to know
- 24:41Main takeaways
- 25:29Questions
- 26:29Applications
related talks
31mBeyond the Black Box: Interpreting ML Models with SHAP
35mSHAPtivating Insights: unravelling blackbox AI models
27mInterpretability and explainability in machine learning
39mExplainable AI as a Conventional Data Analysis Tool
01h 36mOpen the Black Box: an Introduction to Model Interpretability with LIME and SHAP
29mOpeningTheBlackBox
46mExplaining a Machine Learning blackbox
34mRethinking Feature Importance-Evaluating SHAP & TreeSHAP for Tree-Based MLM