16.1 Introduction

Model’s interpretability is crucial for:

• Business adoption
• Model documentation
• Regulatory oversight
• Human acceptance and trust

Interpretable Machine Learning provides solutions to explain at different levels.

• Global Interpretation explain how the model makes predictions, based on a holistic view of its features. They can explain:
  • which features are the most influential (via feature importance)
  • How the most influential variables drive the model output (via feature effects)
• Local Interpretation becomes very important when the effect of most influential model’s features doesn’t the biggest influence for a particular observation as it helps us to understand what features are influencing the predicted response for a given observation or an small group of observations.
  • Local interpretable model-agnostic explanations (LIME)
  • Shapley values
  • Localized step-wise procedures