ISOMAP is a non-linear technique for dimension reduction which allows you to find relationships in an unordered dataset.
What do we mean by non-linear?
As opposed to other linear techniques used for dimension reduction like PCA or SVD, data points that, in appearence, are close to each other (small Euclidean distance) could in fact be farther (Geodesic distance)
In the Swiss Roll picture below, the red and blue dots are very close in terms of the Euclidean distance, but very far in the Geodesic distance.
If you find this interesting, you can take a look at the following link. With the help of R and the “RDRToolbox” package I’ve developed this Cool Shiny.