I have a classification problem and I am doing the data analysis. I came across a variable which is numeric continuous and have some missing values.

I checked the missing values and are real missing values given that they don't exist in the source of information.

My questions is and is a general question, which is the best way to treat these missing values?

I don't like the imputation way (mean,median, etc) because add noise to the model, The only thing that I could think of was to bin the variable and add the missing values as a category, but dont like too much because the loss information in the bin process.

Exist another method?...

This is my first question, please forgive me in advance for any mistake and error.

I didn't provide any data given that is a general question.

Best Regards!

  • $\begingroup$ As for the existence of another method, you could just create your model without the incomplete records. Also, I’ve heard of predicting the missing values using the other predictor variables. I’m not sure the pros/cons of that. $\endgroup$ – Joe Aug 22 '19 at 0:37
  • $\begingroup$ You could try with multiple imputations if you are worried about noise. Or you can add a dummy variable for the missing values and replace the missing values with some uninformative value. $\endgroup$ – user2974951 Aug 22 '19 at 6:10
  • $\begingroup$ Good introductions to this broad topic are Paul D. Allison's book Missing Data and J. W. Graham's article "Missing data analysis: Making it work in the real world" at personal.psu.edu/jxb14/M554/articles/Graham2009.pdf . $\endgroup$ – rolando2 Aug 23 '19 at 4:50

Your Answer

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

Browse other questions tagged or ask your own question.