Hi, I am doing researches about transfer learning for XGboost. I am currently working with a small dataset from a company in Spain (short history) and the scoring is poor. I have worked before with the same company in France and the scoring was great as I had plenty of data thanks to a big history. How could I improve my score with the data from Spain with the help of data from France ? Could I use transfer learning, or data mutualization, or data augmentation ? If anyone has faced before a similar problem, or has read some papers about it, I would love to hear about it. Thank you!
I assume you're dealing with tabular data.
A small dataset might be the cause of the problem, but not always. If you have similar datasets, you could train on one and test on the other. However, if datasets are using different variables, there is nothing you could have done.
Transfer Learning is Okay, however., you need to focus on collecting more data or to use some classical methods, such logestic regression or SVM etc. You may have a look at this link https://www.kaggle.com/code/lucidlenn/data-analysis-and-classification-using-xgboost/notebook too.
Valentyn Bezshapkin Yes those are tabular data. We have already tested training on the bigger dataset and perform on the other, but the scoring was not great.
- Badges
- Science topic