I have a dataset which contain many continuous and categorical features and categorical output. Not all the features are correlated with output, so I took a subset of features - only those which show significant correlation with output.

My question is - While doing outlier detection, should I use all features, or only those which I took as subset (correlated with output)? How will these two approaches affect the results?

  • $\begingroup$ If the output is categorical, how are other variables "correlated" with it? Also, removing outliers could change correlation (or whatever measure you used). $\endgroup$
    – Peter Flom
    Jul 18, 2018 at 11:16
  • $\begingroup$ @PeterFlom I used chisq and Cramer's V to determine the strength of association between categorical features and output. By correlation here I mean strength of association. $\endgroup$
    – Ankit Seth
    Jul 18, 2018 at 11:30
  • $\begingroup$ @PeterFlom Also, I am not measuring strength of association after removing outliers, I am doing outlier detection after measuring strength of association. (I don't know if its the right order of doing it.) $\endgroup$
    – Ankit Seth
    Jul 19, 2018 at 4:56
  • $\begingroup$ Why are you trying to detect outliers? What's the purpose of your overall analysis? $\endgroup$
    – Peter Flom
    Jul 19, 2018 at 10:35
  • $\begingroup$ @PeterFlom I am creating a machine learning model on the data. I created many and got nearly 70% accuracy in all of them, so I thought may be because of outliers I am not getting more accuracy. $\endgroup$
    – Ankit Seth
    Jul 19, 2018 at 11:53


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.