I am working on a project with a sample size of 30. I have 7 features predicting a continuous variable where I am aiming to optimize the r-value. If I change the random seed of my train_test_split() 100 times, I get correlation coefficients that range between .6 and .9 with an occasional negative r-value.

How should I interpret this? Is this variability normal for such a small data set and a random shuffling of my training data? Or is something else going on? How should I decided which seed to use?


This means that your model is not consistent. Your problem is hard because of the small sample size. I would recommend reducing the number of predictor variables.

7 predictors is quite a few for linear regression, even with a large sample size. Try and get the best results you can with 2 or 3 predictors.

If you can predict with 30% accurately all of the time it is better than 70% accuracy 10% of the time.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.