Explainability methods in machine learning

We need to keep the following considerations in mind when using or developing explainability techniques for machine learning modeling (Ribeiro et al., 2016):

• Interpretability: The explanations need to be understandable to users. One of the main objectives of machine learning explanation is to make complex models understandable for users and, if possible, provide actionable information.
• Local fidelity (faithfulness): Capturing the complexity of models so that they are completely faithful and meet global faithfulness criteria can’t be achieved by all techniques. However, an explanation should be at least locally faithful to the model. In other words, an explanation needs to properly explain how the model behaves in the close neighborhood of the data point under investigation.
• Being model-agnostic: Although there are techniques that are designed for specific machine learning methods, such as random forest, they are supposed...