Unsupervised Learning

Abstract

In this chapter, we address the problem of analyzing a set of inputs/data without labels with the goal of finding “interesting patterns” or structures in the data. This type of problem is sometimes called a knowledge discovery problem. Compared to other machine learning problems such as supervised learning, this is a much more open problem, since in general there is no well-defined metric to use and neither there is any specific kind of patterns that we wish to look for. Within unsupervised machine learning, the most common type of problems is the clustering problem; though other problems such as novelty detection, dimensionality reduction and outlier detection are also part of this area. So here we will discuss different clustering methods, compare their advantages and disadvantages, and discuss measures for evaluating their quality. The chapter finishes with a case study using a real data set that analyzes the expenditure of different countries on education.