Using instance-based models for classification and clustering

Instance-based machine learning algorithms are usually easy to understand as they have some geometrical intuition behind them. They can be used to perform different kinds of tasks, including classification, regression, clustering, and anomaly detection.

It's easy to confuse classification and clustering at first. Just to remind you, classification is one of the many types of supervised learning. The task is to predict some discrete label from the set of features (Figure 3.4, left pane). Technically, classification goes in two types: binary (check yes or no), and multiclass (yes/no/maybe/I don't know/can you repeat the question?). But in practice, you can always build a multiclass classifier from several binary classifiers.

On the other hand, clustering is the task of unsupervised learning. This means that, unlike classification, it knows nothing about data labels, and works out clusters of similar samples in your data on its own. In the next chapter, we are going to discuss an instance-based clustering algorithm called k-means (KNN), and in this chapter, we focus on applications of instance-based algorithm KNN to multiclass classification: