Explaining Why You have a Favorite Cereal
summary
Linear regression on a cereal ratings dataset (sourced from kegel.com) yields decent fit but interpretation matters: the model uses raw coefficients that change with units (weight in ounces vs grams) so magnitude is misleading. Post‑hoc SHAP explanations show carbs, sugar and sodium drive higher ratings while a binary LIME check and SHAP waterfall give fruit components a near‑zero attribution; a few example rows (ratings ~64.8, 58.3, 37.1) illustrate local vs global differences and the shared Colab notebook reproduces the results.
AI-written summary · about
chapters (10)
- 0:00Explainable AI introduction
- 3:59Model opacity harms stakeholders
- 7:11Explainability vs interpretability
- 9:16Cereal dataset demo
- 12:49Linear regression needs explanations
- 14:38SHAP values with Game Theory
- 17:07Global explainability with PD plots
- 19:16Local explainability with LIME
- 22:03Fruit components in cereals
- 24:53Summary and Q&A
related talks
35mTaking a peek under the hood - Interpreting black box model
29mDeciphering Sales Drivers at PepsiCo
31mBeyond the Black Box: Interpreting ML Models with SHAP
33mPeople who liked this talk also liked ... Building Recommendation...
27mFast, intuitive feature selection via regression on Shapley values
39mExplainable AI as a Conventional Data Analysis Tool
27mInterpretability and explainability in machine learning
31mEverything in its Right Place: Optimising Ranking in Online Grocery