Dimensionality reduction methods transform an original feature set into a new feature set that usually contains new features that weren't present in the initial dataset. These methods can also be divided into two subclasses—linear and non-linear. The non-linear methods are usually more computationally expensive, so if we have a prior assumption about our feature's data linearity, we can choose the more suitable class of methods at the initial stage.
The following sections will describe the various linear and non-linear methods we can use for dimension reduction.