Explanation

Relates the feature values of an instance to its model prediction in a humanly understandable way.

The answer to a why-question.

An Explainable AI method can give many explanations for a given model. It is not deterministic but should be stable. Only valid for one specific prediction. Explanations make a model more interpretable.

There exist

• Local Explanation
• Global Explanation

Properties

Higher is better.

• Accuracy
  • important when using explanations instead of model predictions (e.g. Decision Tree which approximates a Neural Network)
• Fidelity
• Consistency
• Stability
• Comprehensability
• Cetainty
• Degree of Importance
• Novelty
• Representativeness
• Selectivity

What makes an explanation human-friendly?

See https://christophm.github.io/interpretable-ml-book/explanation.html