What is Unsupervised Learning?

Suppose you enter a party where you don’t know anyone else in attendance. You don’t know who is friends with whom, but you would like to find out based on how they interact with each other. Using clues such as body language, facial expressions and the amount of time people spend talking, you can begin to formulate an idea of the relationships between the party guests—even without being told what these relationships are.

Of course, trying to create sense out of nothing isn’t just a human trait. Given a set of data points, computers can learn the complex relationships between these data points (e.g., which points are in the same category or cluster) without having the answers explicitly available. This approach is known as unsupervised learning.

Knowing how to apply unsupervised learning properly is a valuable data engineering skill. So what is unsupervised learning, and why is unsupervised learning important for machine learning and data science?

What Is the Definition of Unsupervised Learning?

Unsupervised learning is a type of machine learning in which the model does not have access to a labeled training dataset. In other words, the model does not know what the correct output should be given a particular input.

Rather, the model must discover trends and patterns in the data by itself, creating meaning from raw information. The term “unsupervised” refers to the fact that humans do not have to oversee the model by feeding it correct examples. Examples of unsupervised learning techniques include clustering, anomaly detection, dimensional reduction and data compression.

It’s critical for data science beginners to understand the distinction between unsupervised learning and another standard machine learning paradigm: supervised learning. In supervised learning, the model observes a series of labeled training data and makes predictions based on that data. The essential difference between supervised and unsupervised learning is that supervised learning models have access to labeled training data, while unsupervised learning models do not.

Unsupervised Learning: Examples

One of the most common unsupervised learning algorithms is k-means clustering, in which the model partitions the input data into a specified number of clusters (referred to as k). Researchers provide the number k (e.g., 3 or 7) before the task begins. Through a series of repeated steps, the model assigns the input data points into clusters, calculates the center of each cluster and then reassigns the data points based on the closest center.

k-means clustering is practical for many applications, from image compression to pattern recognition. For example, k-means can be used to perform customer segmentation of a company’s audience. The input data consists of various features about each customer (e.g., their age, location, income, etc.). Customers are organized into clusters based on their similarity, which allows businesses to design their marketing efforts to appeal to different customer segments.

Another popular unsupervised learning algorithm is principal component analysis (PCA). PCA is used for high-dimensional datasets in which many of these dimensions are correlated with each other. Machine learning researchers can apply PCA to transform the dataset and reduce its dimensionality while retaining the most useful information and features that separate one data point from the next. PCA is a valuable unsupervised learning technique because it makes information smaller, more manageable and easier to analyze. 

Why Is Unsupervised Learning Important?

Supervised learning is ideal when the problem already has a dataset available for training that has been cleanly separated into labeled data points. Unfortunately, this is not always possible with real-world use cases. Instead, machine learning researchers must often work with raw input data collected without labels.

These are precisely the situations in which unsupervised learning excels. Unlike supervised learning, unsupervised learning can deal with large quantities of unlabeled and unstructured information. In unsupervised learning, the model seeks to learn not the correct labels for the data points themselves but the underlying structure or sense of the distribution from which the data points are derived.

In fact, unsupervised learning is often used as a preprocessing step for supervised learning. First, researchers apply unsupervised learning to the unlabeled input data, separating the data points into predicted clusters and giving each cluster a label. Then, researchers can train a model using supervised learning using the labeled dataset. Knowing how to use unsupervised learning effectively is a crucial data science career skill.