The power of a well-chosen train-test split lies in its ability to answer the right questions.

Try this: Experiment with different train-test ratios to see how it affects your model's performance.