[Audit] - MCAR : Explore the pklm test
adriencrtr opened this issue · 0 comments
adriencrtr commented
Why ?
There are 3 types of missing data : MCAR, MAR and MNAR. In the imputer selection process, we must choose a holes generator. It would be suitable to select a holes generator that reflects the holes present in the dataset. The aim is to provide a tool for characterizing the user's hole type.
Current situation :
No tool available today in Qolmat. The Little's test is implemented but it only handles quantitative features and presents some limits :
- Bad power in high dimension.
- Do not detect the HC.
Improvement sought :
Implement briefly the PKLM test to confirm the published performance by the authors.
See the PKLM Test by Spohn et al.