A model predicts delivery time in minutes. Two candidate models on the same test set: Model X has RMSE 8.1 and R² 0.62; Model Y has RMSE 6.4 and R² 0.76. A naive model that always predicts the mean has RMSE 13.0.
Which statement is correct?
Why did you pick that answer? Two or three sentences. The act of articulating it is what builds the judgment — not the click that follows.
Lower RMSE and higher R² point the same way — they must, since R² = 1 − (model SSE / baseline SSE) on the same data. Model Y's errors are about half the baseline's (6.4 vs 13.0), and R² says it explains 76% of the variance the mean cannot. Both models beat the naive baseline, which is the first bar every regression must clear.
These are test-set numbers; a lower test R² is simply a worse fit. Overfitting shows up as a gap between train and test, not as a low test score.
Accuracy is a classification metric. For a continuous target, RMSE, MAE, MAPE, and R² are the tools.
R² is the share of variance explained, not a hit rate. There is no "correct" for a continuous prediction.