16.1 Introduction

Objective: Prediction

Access to a set of variables, \(x_1,x_2,\dots,x_p\), measured on \(n\) observations, and a response \(y\) also measured on those same \(n\) observations.
Objective is to predict \(y\) using \(x_1,x_2,\dots,x_p\)

In many common situations there are a well-developed set of tools for supervised learning:

Regression and classification
- logistic regression, trees, random forests, bagging, boosting
Clear understanding of how to assess the quality of obtained results
- cross-validation, model fit, etc.

Objective: Description

A set of statistical tools intended for when we have a set of variables, \(x_1,x_2,\dots,x_p\), measured on \(n\) observations.
Objective is to uncover interesting patterns in the measurements of \(x_1,x_2,\dots,x_p\)
- Not interested in prediction because we do not have an associated response variable \(y\)
- Can we discover subgroups among the variables or among the observations?
- Is there an informative way to visualize the data?

Unsupervised learning is, in some ways, more challenging than supervised learning.

Tends to be more subjective.
No simple goal for the analysis, such as prediction of a response
Often hard to assess the results obtained from unsupervised learning methods because there is no universally accepted mechanism for validating the results
No way to check our work because we do not know the true answer—the problem is unsupervised

The importance of reliable unsupervised learning methods is growing in a number of fields:

A cancer researcher might assay gene expression levels in patients with breast cancer to look for subgroups among the breast cancer samples in order to obtain a better understanding of the disease
An online shopping site might try to identify groups of shoppers with similar browsing and purchase histories to target coupons, sales, etc.
A search engine might choose what search results to display to a particular individual based on the click histories of other individuals with similar search patterns.