0

Sorry if the question is naive. I have a problem where i should predict the probability that someone will make a claim given a bunch of predictors. Let's say that the claim can take a range of values between 0-5 and it's highly skewed (gamma distribution).

Could someone tell me what is the benefit of using ML logistic regression over a GLM logit model or vice versa?

Thanks

hippocampus
  • 55
  • 5
  • 1
    How are GLM logistic regression and machine learning logistic regression different? // I’m not convinced that any kind of logistic regression is appropriate here. It sounds like you would want to do an ordinal regression to predict the value of the claim. – Dave Mar 04 '22 at 20:09
  • Does ML mean machine learning or maximum-likelihood here? – Michael Webb Mar 04 '22 at 20:19
  • I said that i want to predict the probability that a claim will occur, not the value of the claim, thus I might convert it to a 0-1 problem. Yeah sorry I'm new to GLM and not sure yet if those two concepts are any different.. So sklern.LogisticRegression is the same as doing glm(family=binomial) from statsmodels? – hippocampus Mar 04 '22 at 20:21
  • @MichaelWebb machine learning – hippocampus Mar 04 '22 at 20:23

1 Answers1

2

There is no "ML logistic regression" there is only one logistic regression. You could see some differences in the results of using logistic regression in, say, R vs scikit-learn because of using different optimization algorithms, also in scikit-learn by default uses regularization, but those are technical details, while the algorithm remains the same.

Tim
  • 108,699
  • 20
  • 212
  • 390
  • I think either sklearn or statsmodels does not automatically use an intercept, too. That could make a huge difference. – Dave Mar 04 '22 at 20:57
  • @Dave I don't use statmodels, but scikit-learn by default uses intercept. It would be a strange and bad default not to use it. – Tim Mar 04 '22 at 21:01