logistic regression in R - Outliers treatment ?

Question

during data exploration , i found lot of outliers and 0 being present (as a value) in data set , for instance freq of 0 is 377 out of 789 observations . do i need to treat these outliers for better accuracy or not ?

Answer 1

It depends on what you are trying to do: Which question you want to answer, what assumptions you have about the data...

One does not remove outliers just to make it look "good". You have to give convoncing arguments if you exclude data. For instance, do you have reasons to believe that these "outliers" are invalid data? Depending on the data (and their distribution) different techniques of outlier detection might be appropriate.

Remember: Not only leaving outliers in the data but also removing alleged outliers can bias your result and conclusion. Sometimes outliers are in fact very informative.

Whatever you do, make your analysis transparent and explain and justify your approach.