Byte-Sized ML Series: Feature Importance with SHAP Values

Description:

This 90-minute session introduces participants to the critical topic of model interpretability, focusing on feature importance using SHAP (Shapley Additive explanations). SHAP values provide consistent, model-agnostic explanations for both regression and classification problems. Participants will learn to visualize and interpret feature contributions, enabling them to trust, debug, and communicate model decisions more effectively.

Duration: 90 mins

Course Code: BDT491

Learning Objectives:

After this course, you will be able to:

• Understand the concept of feature importance and why it matters in ML
• Explain how SHAP values quantify individual feature contributions
• Apply SHAP to both regression and classification models
• Visualize and interpret SHAP summary and dependence plots
• Use SHAP to uncover hidden patterns and model biases

• Must have some python programming experience.

• Aspiring and practicing data analysts or ML beginners who have trained basic models and now want to interpret and explain why a model is making predictions. Some familiarity with scikit-learn, regression/classification, and data manipulation with Pandas is expected.

Course Outline:

1. Why Feature Importance Matters?
   • Motivation: trust, transparency, accountability in ML
   • Use cases in business, healthcare, finance
   • Different types of feature importance (built-in, permutation, SHAP)

2. Introduction to SHAP Values
   • What are SHAP values? (Game theory roots simplified)
   • Benefits over traditional feature importance
   • Model-agnostic vs. model-specific SHAP
   • Visual intuition with simple examples
   • Hands-on: Visualize SHAP values
3. SHAP for Regression
   • Load and train a regression model (e.g., RandomForestRegressor)
   • Use “shap.Explainer” to compute SHAP values
   • Visualize: Summary Plot, Waterfall, etc.
   • Hands-on: Interpret feature effects on housing dataset
4. SHAP for Classification
   • Train a classifier (e.g., Random Forest Classifier or XGBoost)
   • Compute and visualize SHAP values for classification
   • Class-wise explanations: understanding prediction drivers
   • Feature importance and model interpretability
   • Hands-on: Explore feature impacts on classification decisions

Training material provided: Yes (Digital format)

Hands-on Lab: Instructions will be provided to install Jupyter notebook and other required python libraries. Students can opt to use ‘Google Colaboratory’ if they do not want to install these tools