Model screening using JMP16 and Imbalanced data

[#1]

Hello

I have an imbalanced dataset with the outcome of interest of only 13% and many predictors. I want to analyse it using the Model screening platform in predictive analysis of JMP 16. In the previous version of JMP15 I was able to use the Add-inn for imbalanced data to look for the best preprocessing technique for my imbalanced dataset. How do I perform this in the present model screening module of JMP16.1? Do I first have to create a new table with  using the Add-inn balancing technique (e.g. SMOTE-Tomek) before starting the analysis? Indeed, preprocessing your imbalanced data can signficantly improve performance of the metrics of your models (see manuscript enclosed)

Looking forward for suggestions

Lu

Belgium

[#2]

Hi,

I am sorry to see that you have not had a reply yet.

I am not able to comment on SMOTE or other pre-processing methods outside of JMP. I would say that it sounds like it could be worth trying.

In JMP Pro I would recommend that you look at the option to create a validation column that is stratified by your response variable. This ensures that the responses levels are equally represented in your test and validation subsets. You can then use this validation column in Model Screening in JMP Pro 16. 

Here is a quick video on how to create the validation column.

Let me know if that helps,

Phil