I am doing a regression problem on a target variable that its values range between 0 to 0.6 and highly skewed between 0 and 0.4.

So, when I use rmse loss, I get the best rmse score as 0.21, is that very bad for the data? What would be a good range for rmse loss for this target variable prediction? and in a DNN-based regression, what are some ways you’d suggest to improve upon the rmse score?

I am doing a random train_test_split using sklearn with test data being 30% of the entire data.

I wanted to do stratify on target variable, but many target values have only 1 and that method requires more than 2 in each bin.

Also, I only have 389 samples in my dataset.

Thanks a lot for any feedback.