One train/test split can be lucky. Cross-validation asks the same question five times so you can trust the answer, with a few traps for time series and grouped data.
No posts here yet.