Chapter 2: Model Explainability Methods
One of the key goals of this book is to empower its readers to design Explainable ML systems that can be used in production to solve critical business problems. For a robust Explainable ML system, explainability can be provided in multiple ways depending on the type of problem and the type of data used. Providing explainability for structured tabular data is relatively human-friendly compared to unstructured data such as images and text, as image or text data is more complex with less interpretable granular features.
There are different ways to add explainability to ML models, for instance, by extracting information about the data or the model (knowledge extraction), using effective visualizations to justify the prediction outcomes (result visualization), identifying dominant features in the training data and analyzing its effect on the model predictions (influence-based methods), or by comparing model outcomes with known scenarios or situations...