Interpreting Black-Box Models with Applications to Healthcare
summary
Toolkit pre-computes model outputs by mutating each feature across its range and produces feature dependence plots, a feature effect summary that ranks variables by median peak-to-trough impact (Ames housing: median living-area effect ≈ $120k on price), and an explain prediction difference routine that greedily attributes shifts (example: ICU MIMIC: 59,000 patients, respiration rate and BUN drive ~10 percentage points of mortality risk on average). Precomputation scales with rows × features × resolution and takes minutes for thousands of rows.
AI-written summary · about
chapters (13)
- 0:00Interpreting Black-Box Models with Applications to Healthcare
- 0:37In The Beginning...
- 1:01Variables had Clear Interpretations
- 1:36Then came Advanced ML
- 2:54Linear / Logistic Regression
- 4:06Three Reasons to Interpret Models
- 7:25Some Work on Model Interpretation
- 11:22ML-Insights Package (Lucena/Sampath)
- 14:49ML-Insights Example 1
- 15:36ML-Insights Example: Housing
- 27:11ML-Insights Example 2
- 28:43ML-Insights Package: Medical
- 35:26Where to Find More
related talks
26mStop Explaining Black Box Models, Use Interpretable Models Instead
29mOpeningTheBlackBox
35mTaking a peek under the hood - Interpreting black box model
15mInterpretable Machine Learning with R
47mBreaking Black-Box AI
22mOne-click analysis: Detecting and visualizing insights automatically
20mDiagnostic Medical application of ML
01h 36mOpen the Black Box: an Introduction to Model Interpretability with LIME and SHAP