Linear regression predicts a numeric target with an intercept and a weighted sum of features. In the fictional delivery records, distance, item count and weekend status explain part of the duration. The model estimates their weights from training rows. Predictions remain numeric and can fall between observed target values.
Fit the estimator before calling predict, then pass feature columns in the same order. A good fit on training rows doesn't remove the need for held-out error. MAE summarizes typical absolute error in duration units; RMSE penalizes large residuals more heavily. An intercept matters because a route can take time even when its numeric inputs are zero. These exercises progress from fitting to error reporting, with fresh records rather than a copied benchmark dataset.