The term unsupervised learning refers to statistical methods that extract meaning from data without training a model on labeled data (data where an outcome of interest is known). In Chapters 4 and 5, the goal is to build a model (set of rules) to predict a response from a set of predictor variables. Unsupervised learning also constructs a model of the data, but does not distinguish between a response variable and predictor variables.
Unsupervised learning can have different possible goals. In some cases, it can be used to create a predictive rule in the absence of a labeled response. Clustering methods can be used to identify meaningful groups of data. For example, using the web clicks and demographic data of a user on a website, we may be able to group together different types of users. The website could then be personalized to these different types.
In other cases, the goal may be to reduce the dimension of the data to a more manageable set of variables. This reduced set could then be used as input into a predictive model, such as regression or classification. For example, we may have thousands of sensors to monitor an industrial process. By reducing the data to a smaller set of features, we may be able to build a more powerful and interpretable model to predict process failure than by including data streams from thousands of sensors.
Finally, unsupervised learning can be viewed as an extension of the exploratory data analysis (see Chapter 1 ...