Unsupervised Learning

What is unsupervised learning?

In supervised learning, we discussed that the models (or classifiers) are built after training data, and attributes are linked to the target attribute (or label). These models are then used to predict the values of the class attribute in test or future data instances. Unsupervised learning, however, is different. There’s no target or class attribute. The methods of unsupervised learning are used to find underlying patterns in data and are often used in exploratory data analysis.

In unsupervised learning, the data is not labeled. The methods instead focus on the data’s features. The overall goal of the methods is to find relationships within the data and group data points based on some similarity matrix.

Some common methods and techniques used in unsupervised learning include the following.

Clustering

The simplest and most popular use of unsupervised learning is clustering. Finding clusters, or subgroups within unlabeled data, is the goal of clustering. Data points in a cluster will be as similar to one another as feasible and as different from data points in other clusters as feasible. Clustering makes it easier to see underlying patterns in the data that a human observer would miss. Clustering algorithms can be divided into two main categories.

Partitional clustering

The goal of partitional clustering is to divide data points into discrete, nonoverlapping groups. The objective is to maximize a certain criterion, like minimizing the sum of squared distances within clusters, and each point is assigned to a single cluster. The partitional clustering algorithm can further be divided into two types. They are discussed below.

The $k$-means clustering algorithm is one of the most commonly used clustering algorithms. It divides the given data into $k$ clusters. The algorithm expects  ...