t-SNE and UMAP for Dimensionality Reduction — Data Analysis & Feature Engineering

t-SNE and UMAP are non-linear dimensionality reduction methods that create two- or three-dimensional visualisations preserving the local structure of high-dimensional data, revealing clusters, gradients, and separations that PCA's linear compression cannot show. They exist because most real data does not lie on a flat plane in high-dimensional space but on a curved manifold — the intrinsic geometry of a face or a cricket player's skill profile is not linear — and linear methods like PCA can only unfold flat structure. t-SNE focuses on preserving local neighbourhood relationships, producing beautiful cluster visualisations, while UMAP is faster, scales to larger data, and better preserves global structure. Together they are the standard tools for exploratory visualisation of complex high-dimensional datasets, helping analysts understand structure before modelling and diagnose whether classes are separable.

Analogy🏏Cricket

🏏 Think of it like cricket: Reducing twenty batting statistics to two dimensions with PCA produces a flat projection that may look like one undifferentiated cloud, while t-SNE or UMAP finds that the data actually organises into distinct clusters — aggressive pinch-hitters, steady anchors, explosive finishers — whose separation PCA's flat projection smeared together. Just as the non-linear reduction reveals the natural groupings that flat projection could not, t-SNE and UMAP reveal structure that PCA misses because it can only flatten, not curve to follow the data's natural shape. The insight is that non-linear reduction methods follow the data's true curved geometry rather than forcing a flat projection, revealing the cluster structure and local neighbourhoods that linear methods like PCA cannot preserve.