Also known as dimensionality reduction.
The aim of PCA is to create new features from the existing features which describes our data well.
Let us suppose we have 1000 features. Many of them can be correlated. So, to minimize to the amount of data, we will create some 100 features from those exisiting 1000 features which will describe aur data well.
- Speed
- Memory
- Visualization ( 2D, 3D data can be easily visualized)
- We will lose some information after applying it. But we minimize the loss of info by creating the best feature.
Feature scaling for PCA is must. If we do not feature scale then PCA will end up choosing the wrong feature.
For eg: