This is perhaps the most important chapter. The fundamental question addressed in this chapter is as follows:
- How do we select a model that predicts well?
This is the purpose of cross-validation, regardless of what the model is. This is slightly different from traditional statistics, which is perhaps more concerned with how we understand a phenomenon better. (Why would I limit my quest for understanding? Well, because there is more and more data, we cannot necessarily look at it all, reflect upon it, and create a theoretical model.)
Machine learning is concerned with prediction and how a machine learning algorithm processes new unseen data and arrives at predictions. Even if it does not seem like traditional statistics, you can use interpretation and domain understanding to create new columns (features) and make even better predictions. You can use traditional statistics...
