-1
$\begingroup$

I have about 100 data sets with over 100,000 rows and 27 predictor variables and multiple response variable on each dataset. I want to select the most essential Features for these datasets. I used Boruta() from the package Boruta, but it takes an extremely long time (about 3 hours for a single sample). Could someone kindly tell me the quickest technique to select key features from a large dataset?

This is the extension from my previous questions:

  1. https://stackoverflow.com/questions/76824529/how-to-do-regression-modelling-in-r-multiple-one-at-a-time-to-multiple-column
  2. How to get coefficient tables from multiple regression model result!
Improve this question
$\endgroup$
2
  • $\begingroup$ Your second question has nothing to do with machine learning, it's a simple R question - you ran into problem because you did not use the apply family of functions properly. $\endgroup$
    – Ram RS
    Commented Aug 16, 2023 at 19:39
  • $\begingroup$ Right, I thought that too. Really an analyst uses either ML or GLM - not both. Within ML it's common to compare different methods but not move outside ML (unless they want to do deep learning). Its possible "linear regression" ML might be intended. ML is very, very different to GLM, i.e. the method and system to test model. $\endgroup$
    – M__ ♦
    Commented Aug 16, 2023 at 20:06

1 Answer 1

Reset to default
0
$\begingroup$

Feature selection: the fastest thing moving (i.e. the simplest) is naive Bayes ML. I don't use it but KNN (K-nearest neighbor) I do use, which is also computationally simple, i.e. fast. I think it gives better results, i.e. the best results for a simple algorithm.

R ML has improved decision trees BTW - quite popular, but you'd never use this in Python.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.