Chapter 6 – Interpretability and Explainability in Machine Learning Modeling
- Explainability can help improve performance, such as by reducing the sensitivity of models to small feature value changes, increasing data efficiency in model training, trying to help in proper reasoning in models, and avoiding spurious correlations.
- Local explainability helps us understand the behavior of a model close to a data point in feature space. Although these models meet local fidelity criteria, features that have been identified to be locally important might not be globally important, and vice versa.
Global explainability techniques try to go beyond local explainability and provide global explanations to the models.
- Linear models, although interpretable, usually have low performance. Instead, we could benefit from more complex models, with higher performance, and use explainability techniques to understand how the model comes up with its predictions.
- Yes, it does...
