1
$\begingroup$

I am trying to use basic classification algorithms in Python using sklearn. Unfortunately my dependent variable in training data are probabilities of whether it will rain or not. I know how to do this if my dependent variable is "rain" or "sunny". But now it is a probability. Can I use sklearn classification algorithms for this problem or should I use regression? Please advise.

$\endgroup$
3
  • 1
    $\begingroup$ After googling a bit, I realized that I could convert the [0,1] probability into a normally distributed variable using log (1 / (1 - x)) transformation and then run the regressions. I am still unsure how to solve this as a classification problem. $\endgroup$ – pavybez Dec 23 '16 at 11:36
  • 1
    $\begingroup$ If you have the base counts the probabilities came from (the denominator of the prob calculation), you can easily alter data to use classification algorithms. $\endgroup$ – B_Miner Jan 12 '17 at 15:55
  • 1
    $\begingroup$ Otherwise you could try what you suggest, but Beta regression is developed specifically for this type of problem. $\endgroup$ – B_Miner Jan 12 '17 at 15:57
0
$\begingroup$

I do not see any chance to model a continuous output between [0,1] as a classification problem.

Use regression with P('Sunny') as the target variable.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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