I am building a classification model using Random Forrest, Decision Trees & Logistic regression.
For one of my critical variable i have some missing values. In some earlier models during similar situation i have used mean of that variable for the missing entries.
I see that few analysts use median or even a 0 for the missing value. What is the rational behind it and does it vary from one use to another.
Thank you.
Hi Shivi,
People tend to apply several approaches to deal with missing data. Replacing missing values with a central tendency for the attribute", e.g., the mean or median, is one of those approaches. Another approach is using 0 to indicate that no value has existed for the specific instance. The results differ from case to another based on the characteristics of the utilized data.
HTH.
Samer
In my opinion, it isn't fully correct to insert mean value instead of missing value. Some parameters may be strong correlated, and in this case to insert missing value of some parameter we should take into account values of other (not missing!) parameters and values of correlations. So, you should use expressions for conditional mean value for multi-dimensional normal distribution.
This is good reference for that;
https://www.yorksj.ac.uk/media/content-assets/schools/psychological-social-sciences/documents/How-to-enter-missing-data-in-SPSS.pdf
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification