The regression model is built on log transformed data and later re-transformed .So to validate the model i m using k-fold cross validation method using weka. Obviously here i m getting validation model on log transformed data. My question is can we perform k-fold cross validation with re-transformed model?
As far as I understand, after you got some response values with your regression model, you want to use their re-transformed values for cross validation. For me, it is not proper to do so. Your model is built on transformed data, and your evaluation must proceed with those.
I think you should perform cross validation on transformed data set. Because your model build on the transformed dataset. So testing should be done the transformed dataset.
Hazar Altınbaş sir you are right on first point..that's what I think...but I don't know how it can be done?
Pankaj Das sir this is what I did by performing cross validation on transformed data and accuracy measures like MMRE ,PRED(25) are too high..Even the reviews of my paper are suspicious about accuracy and asked if it is performed on trnasformed or re-transformed dataset. Any help?
First transform data then build model on the basis of transformed data. And check validation.
Another option is that first clean data specially the extreme values of data. Then fit model on the cleaned dataset
Pankaj Das sir thats what I did..on transformed data..n validation also. But reviewers say otherwise.
Split dataset into training and testing set in 80:20 ratio. Train your model on the 80% data ( training data set). Then use remaining 20% data (test data) for testing the model.
Do the cross-validation with transformed data to be sure about robusteness of your model. After, you have your log type responses (as predictions obviously) use exponential function to inverse log transformation.
Hazar Altınbaş thank you..I have done the same. The problem was with calculation of MRE for log transformed data..I have found the answer..thanks for support.
Anureet Kaur Can you please point me to the answer you found? I seem to have the same dilemma here. I can do cross-validation on the log-transformed data (independent variable log-transformed), but I have several models to compare, some using log-transformed data, some not. It is obviously not possible to simply inversely transform the log-model validated MSE and RMSE, since it gives incomparable results vis a vis the models without transformation, which suggests such inversely transformed MSE and RMSE are invalid.
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Publication