Hi there,
you want to make sure that your data, considering the chosen split is representative of your data and real world problem. E.g. your training set should contain all representative conditions so that your model has the chance to learn all relevant characteristics from the data.
In case this applies here, I assume your friend and you would end up with comparable metrics (with respect to R^2) given that all other (hyper-) parameters and boundary conditions are being equal. I think it would be useful to do also a residuum analysis and check the distribution.
Feel free to to this comparison and interpret or discuss your findings.
A nice way to get familiar with different splits (not necessarily split ratios) is cross validation.
It’s can also be helpful for dealing with overfitting issues.
Feel free to take a look:
- Cross-validation (statistics) - Wikipedia
- 3.1. Cross-validation: evaluating estimator performance — scikit-learn 1.1.1 documentation
Best regards
Christian