Summary

In this chapter, we learned about classification using k-NN. Unlike many classification algorithms, k-nearest neighbors does not do any learning—at least not according to the formal definition of machine learning. Instead, it simply stores the training data verbatim. Unlabeled test examples are then matched to the most similar records in the training set using a distance function, and the unlabeled example is assigned the label of its nearest neighbors.

Although k-NN is a very simple algorithm, it can tackle extremely complex tasks, such as the identification of cancerous masses. In a few simple lines of R code, we were able to correctly identify whether a mass was malignant or benign 98 percent of the time in an example using real-world data. Although this teaching dataset was designed to streamline the process of building a model, the exercise demonstrated the ability of learning algorithms to make accurate predictions much like a human can.

In the next chapter...