How to fit a multiple linear regression model on 1664 explanatory variables in R?

So we need to choose from the 1664 variables a legitimate model, for example a model that predicts as a significant part of the inconsistency in the information with as scarcely any logical variables. There are a few different ways of doing this:

Utilizing master information to choose variables that are known to be significant. This can be because of different examinations discovering this, or because of some fundamental procedure that we presently make that variable applicable.

Utilizing a stepwise regression approach which chooses the variables are applicable depending on how well they clarify the information? Do take note of that this technique has some genuine drawbacks. View stepAIC for a method for doing this utilizing the Akaike Information Criterion.

Connecting 1664 variables with information will yield around 83 critical relationships in the event that we pick a 95% importance level (0.05 * 1664) absolutely dependent on irregularity. Thus, track cautiously with the programmed variable choice. Chopping down the measure of variables with master information or some decorrelation procedures (for example head part examination) would help.