How t-SNE Outperforms PCA in Dimensionality Reduction

PCA vs t-SNE for visualizing high-dimensional data in a lower-dimensional space

Photo by Pat Whelen on Unsplash

In machine learning, dimensionality reduction refers to reducing the number of input variables in the dataset. The number of input variables refers to the dimensionality of the dataset.

Dimensionality reduction techniques are mainly divided into two main categories: Linear and Non-linear (Manifold).

Under linear methods, we have discussed Principal Component Analysis (PCA), Factor Analysis (FA), Linear Discriminant Analysis (LDA) and Non-Negative Matrix Factorization (NMF).

Under non-linear methods, we have discussed Autoencoders (AEs) and Kernel PCA.

t-Distributed Stochastic Neighbor Embedding (t-SNE) is also a non-linear dimensionality reduction method used for visualizing high-dimensional data in a lower-dimensional space to find important clusters or groups in the data.

All dimensionality reduction techniques fall under the category of unsupervised machine learning in which we can reveal hidden patterns and important relationships in the data without requiring labels.

So, dimensionality reduction algorithms deal with unlabeled data. When training such algorithms, the…