Model Evaluation

Performance Metrics

Comprehensive evaluation showing R² of 0.840, RMSE of £1,900, and stable cross-validation

Metrics

Evaluation Metrics

All metrics computed on the held-out test set (33% of data)

£1,400
Mean Absolute Error
£1,900
Root Mean Squared Error
0.840
R² Score
0.839
Adjusted R²
0.840
CV Mean R²
< 0.01
CV Std Deviation
Understanding

What Do These Metrics Mean?

MAE (Mean Absolute Error)

£1,400 — On average, predictions are off by £1,400 from the actual price.

This is the most intuitive error metric. Lower is better.

RMSE (Root Mean Squared Error)

£1,900 — Typical prediction error, with more penalty for large mistakes.

Squares errors before averaging, so large errors are weighted more heavily.

R² (R-squared)

0.840 — The model explains 84% of the variance in price.

Range: 0 to 1. Higher is better. 0.80+ is considered excellent for regression.

Adjusted R²

0.839 — R² adjusted for the number of features in the model.

Penalises adding too many features. Very close to R², indicating good feature selection.

CV Mean R²

0.840 — Average R² across 5 different train/test splits.

Cross-validation tests how well the model generalises to unseen data.

CV Std Deviation

< 0.01 — Very low variance across CV folds.

Low std means the model is stable and not sensitive to the data split.

Performance Summary

Strengths

  • Exceeds target: R² of 0.840 surpasses the 0.80 goal
  • Stable: CV standard deviation < 0.01 confirms generalization
  • Interpretable: Coefficients show clear feature impact
  • Fast: Training < 100ms, inference < 1ms per prediction
  • Production-ready: Complete reproducible pipeline

Limitations

  • Linear assumption: Relationship between mileage and price is partly non-linear
  • Extremes: Performs less well on very cheap or very expensive vehicles
  • Outliers: Doesn't clean data quality issues (year=2060, MPG>150)
  • Hybrid/Electric: Limited training data for these premium fuel types

Business Interpretation

A dealer using this model can expect predictions within approximately £1,900 of the true market price for a typical Ford vehicle. The model provides a consistent, objective baseline that removes human guesswork from initial appraisals. Hybrid and Electric vehicles should be reviewed manually as the model has limited training data for these categories.

Visualizations

Performance Graphs

Click any graph to view full screen

Loading visualizations...

5-Fold Cross Validation

To ensure the model isn't overfitting to a particular train/test split, we performed 5-Fold Cross Validation. The dataset is split 5 times, and the model is trained and evaluated on each split.

Fold R-squared Status
Fold 1 ~0.840 Excellent
Fold 2 ~0.840 Excellent
Fold 3 ~0.841 Excellent
Fold 4 ~0.839 Excellent
Fold 5 ~0.840 Excellent
Mean ~0.840 Very Stable
Std Deviation < 0.01 Excellent

Verdict

The extremely low standard deviation (< 0.01) across all 5 folds confirms the model generalises well and is not overfitting to any particular data split. This is a very strong indicator of model stability.

Ready to Make Predictions?

Try the interactive prediction tool or explore the key insights from the project