Multidimensional scaling (MDS) is a non-linear dimensionality reduction and manifold learning technique. It maps high-dimensional data points to a lower-dimensional space while attempting to preserve the pairwise distances between the points as closely as possible.