Unsupervised Machine Learning is a type of learning where there is no labeled data. The objective of unsupervised learning is for the learning algorithm to find patterns by similarities or differences from the data. Unsupervised learning systems are often used in tasks such as segmentation of customer data, grouping of political supporters and detecting hackers attacks among many other use cases. In this post we are going to look at unsupervised machine learning how it works, its challenges, strengths and limitation.

Unsupervised Machine Learning

unsupervised learning - unsupervised machine learning
Image from http://plus.usgbc.org/from-polluted-metropolis-to-forest-city/

As the name suggests, in unsupervised machine learning there is no target variable, therefore the algorithm has to “blindly” find patterns from the data since there is no prior training/knowledge. Despite its wide application, it is not easy to evaluate the output of unsupervised learning model. However, unsupervised machine learning models are very useful when dealing with complex tasks that are hard to solve by supervised learning models.

Types of Unsupervised Learning Algorithms.

There are many unsupervised learning algorithms grouped into different categories depending on there applications;

  • Clustering. These are the most commonly used unsupervised learning algorithms. Clustering algorithms are used for grouping data with similar pattern. Clustering algorithms includes k-means and hierarchical algorithms.
  • Neural Networks. Artificial neural networks (ANNs) have revolutionized the machine learning space with its power to learn from the unseen patterns in the data. Unsupervised algorithms in this category includes; Genarative Adversarial Networks, Deep Belief Networks, Self-Organizing Map and Autoencoders among others.
  • Anomaly Detection. These are sets of algorithms that are used in identification of events that are unusual in a data set such as fraud and errors in systems. Algorithms in this category includes; self-organizing maps (SOM), K-means, C-means, expectation-maximization meta-algorithm (EM) and one-class support vector machine.
  • Dimensionality Reduction. These are set of algorithms that performs data compression. Dimensionality reduction makes data small and more efficient by removing unnecessary (redundant) information. This is done by eliminating some features in your data. The algorithms in this category includes; Principal Component Analysis (PCA), Independent Component Analysis and Singular Value Decomposition (SVD).

Applications of Unsupervised Machine Learning

Unsupervised learning model are used together with supervised learning models in applications such as;

  • Self-driving cars
  • Expert systems
  • Robotics
  • Image Processing.
  • Density estimation
  • Customer segmentation

Challenges With Implementing Unsupervised Learning

  • It is difficult to measure the accuracy of unsupervised learning model. Perhaps with more research effort being done this will not be an issue soon.
  • There is no any clear point on when we need to stop when using the hierarchical clustering algorithms.
  • Difficulty in selecting the right similarity distance such as Euclidean, cosine e.t.c.


Unsupervised machine learning is becoming more important day by day with many tasks that can only be solved by such algorithms. Unsupervised learning plays an important role in machine learning and AI in tasks such as dimensionalty reduction and testing other AI systems. One of the downside of unsupervised learning is the inability to evaluate its accuracy.

What’s Next

In this post we have introduced ourselves to the unsupervised machine learning. In the coming series of post we are going to look at the most commonly used unsupervised learning algorithms starting with the clustering algorithms.

Unsupervised Machine Learning

Post navigation