I am working on a prediction problem that leverage sparse clinical datasets. Missing data rate is in the range of 80%.

1- I am wondering if there is any example of application of matrix completion to clinical or other datasets with such a missing rates.

2- Currenlty exploring glmnet, pcaMethods and SoftImpute pacakages. I am also looking for R packages/SAS routines that can handle such sparse clinical data matrix and perform matrix completion.

3- I would like assess the reliability of my filled-in values, is there any metric or score to assess the quality of the matrix completion.

Thanks in advance !

  • $\begingroup$ Here is a resource that also points to some other potential approaches for you: (Huang, J., Nie, F., & Huang, H., 2013). $\endgroup$ – jvbraun Dec 1 '14 at 8:04
  • $\begingroup$ Thanks jvbraun. I sorted this issue using a combination of labelled value imputation (high, very-high, extremely-high, normal, low, very-low and extremly low) and real-value imputation. Tested using a bunch of different methods and PCA-based methods designed to handle healthcare / biomedical data did the trick at the end. $\endgroup$ – Khader Shameer Dec 13 '14 at 23:39

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.