Tuning models with cross-validation

We can simply avoid adopting the classification results from one fixed testing set, which we did in experiments previously. Instead, we usually apply the k-fold cross-validation technique to assess how a model will generally perform in practice.

In the k-fold cross-validation setting, the original data is first randomly divided into k equal-sized subsets, in which class proportion is often preserved. Each of these k subsets is then successively retained as the testing set for evaluating the model. During each trial, the rest of the k -1 subsets (excluding the one-fold holdout) form the training set for driving the model. Finally, the average performance across all k trials is calculated to generate an overall result:

Figure 2.9: Diagram of 3-fold cross-validation

Statistically, the averaged performance of k-fold cross-validation is a better estimate of how a model performs in general. Given...