I just watched the bit about Principal Component Analysis. In the lecture Andrew says PCA is mainly used for visualisation, which I found odd.
In my mind, the purpose of PCA was to reduce the issue of multicollinearity related to trying to train a model with too many features that are linearly related related somehow (e.g. GDP, GDP per capita, etc.).
Is my understanding wrong?
You mentioned watching a video about PCA. I assume this is part of a course you are taking currently. Which course is it? Try moving your topic to the course and the associated week so that the assigned mentors for the course can do justice to your question.
What course are you attending?
FYI, PCA is discussed in MLS Course 3 Week 2 as an Optional topic.
One lecture discusses reducing the number of features, and there is a lab that discusses a visualization application.
Dimensionality reduction is not nearly as popular as it once was, because computers now have lower cost, higher performance, and greater memory capacity. If you have a big enough computer to use all of the original features, there isn’t much advantage in reducing them.