Glossary

Dimensionality reduction

Techniques that compress a dataset with many variables into fewer dimensions while preserving as much meaningful information as possible.

Dimensionality reduction is a set of techniques that compress a dataset with many variables, dimensions, into a smaller set that still captures most of the meaningful information. High-dimensional data can make models slower to train, harder to visualize, and prone to the "curse of dimensionality," where the amount of data needed to reliably learn patterns grows rapidly as more variables are added.

Principal component analysis, the most common method, finds new combined variables, called components, that capture the most variance in the data, ordered so the first few components summarize most of the original information; t-SNE and UMAP are popular alternatives, mainly used for visualizing high-dimensional data in two or three dimensions rather than for feeding into a model. This is a form of unsupervised learning, since it uses no target label, only the structure of the inputs themselves.

Dimensionality reduction is used as a preprocessing step before clustering such as k-means clustering, to reduce noise and speed up algorithms sensitive to many features, and as a feature engineering technique to combine correlated variables into fewer, more informative ones. Its main pitfall is interpretability: reduced components are mathematical combinations of the original variables and often do not have a clean, human-readable meaning, which makes results harder to explain to a non-technical audience than the original features.

Last reviewed September 22, 2026

In the index now

Related terms

Feature engineering Creating, transforming or selecting the input variables a model is trained on, to make the underlying pattern easier for it to learn. Clustering An unsupervised technique that groups data points so items within a group are more similar to each other than to items outside it. Unsupervised learning A machine learning approach that finds structure or patterns in data that has no labeled correct answer to learn from. K-means clustering A clustering algorithm that partitions data into k groups by repeatedly assigning points to the nearest of k cluster centers.