Working with Model Explainability

The justification of model selection and performance is just as important as model training. You can have N trained models using different algorithms, and all of them will be able to make good enough predictions for real-world problems. So, how do you select one of them to be used in your production services, and how do you justify to your stakeholders that your chosen model is better than the others, even though all the other models were also able to make accurate predictions to some degree? One answer is performance metrics, but as we saw in the previous chapter, there are plenty of performance metrics and all of them measure different types of performance. Choosing the correct performance metric boils down to the context of your ML problem. What else can we use that will help us choose the right model and also further help us in justifying this selection?

The answer to that is visual graphs. Human beings are visual creatures and, as such, a...