After this chapter, you should be able to
- Build linear and nonlinear regression baselines.
- Calculate MAE and RMSE in physical units.
- Inspect residual bias and critical slices.
- Distinguish interpolation from extrapolation.
- State why prediction cannot approve material or design.
Engineering context and methodSource §Lesson 04 · Engineering context and method · GOV-01 · RES-01 · RES-03 · RES-04 · DATA-01
Regression predicts a continuous target. Evaluate against a simple mean or linear baseline; retain physical units; plot residuals against predictions, age, material ranges and acquisition groups. MAE communicates typical absolute error while RMSE weights large errors more strongly. Neither metric proves code compliance, causality or safe extrapolation. The UCI concrete dataset is a teaching dataset with 1,030 records, eight inputs and one strength target—not a material-acceptance procedure. In geotechnics, sparse site investigation, stratigraphy and spatial dependence make transfer especially hazardous; group validation by site or geological unit and inspect tail behaviour. Quantity and cost forecasts must retain measurement rules, price date, currency, location and scope, and cannot certify a bill or budget.
Verified worked exampleSource §Lesson 04 · Verified worked example · GOV-01 · RES-01 · RES-03 · RES-04 · DATA-01
MAE and RMSE for strength predictions
Prediction errors are −2, 4 and 1 MPa.
- MAE
(|−2| + |4| + |1|) / 3
7/3 = 2.333 MPa - MSE
(4 + 16 + 1) / 3
7 MPa2 - RMSE
√7
2.646 MPa
Result. MAE = 2.333 MPa and RMSE = 2.646 MPa. Three errors illustrate arithmetic only; they do not validate a model.
Practical lab · 6 h lesson effortSource §Lesson 04 · Practical lab · 6 h lesson effort · GOV-01 · RES-01 · RES-03 · RES-04 · DATA-01
- Use a frozen UCI Concrete release and reproduce its 1,030-row, nine-variable schema.
- Compare mean, linear and tree-based regressors on a fixed independent split.
- Report MAE, RMSE, residual plots, input bounds and error slices by age and strength range.
- Add a short transfer-risk comparison for a site-grouped geotechnical or date/location-controlled quantity-cost use, then write the model card.
Failure modes to investigateSource §Lesson 04 · Failure modes to investigate · GOV-01 · RES-01 · RES-03 · RES-04 · DATA-01
- Specimen or batch dependence ignored.
- Strength age used incorrectly or unit labels dropped.
- Only R² reported.
- Extrapolation beyond observed mix ranges.
- Prediction described as acceptance, conformity or design strength.
Knowledge checksSource §Lesson 04 · Knowledge checks · GOV-01 · RES-01 · RES-03 · RES-04 · DATA-01
| Question | Answer rationale |
|---|---|
| Why report units with error? | Engineering meaning depends on the physical scale of error. |
| Why can RMSE exceed MAE? | Squaring gives larger errors more weight. |
| Does R² measure acceptance risk? | No; it is a variance-explanation measure under the evaluated data. |
| What is extrapolation? | Prediction outside the range or conditions supported by training and validation evidence. |
| Can this model approve concrete? | No; sampling, testing, specification and competent acceptance procedures govern. |
Key points
- Start from the accountable engineering decision and its consequence.
- Compare against a transparent non-AI baseline.
- Validate on a split that represents intended use and retain human authority.
Source references recorded by the supplied chapter
- UCI Concrete Compressive Strength dataset, DOI 10.24432/C5PK67.
- Yaghoubi et al. and Zhang et al., geotechnical ML and reliability reviews recorded as RES-03 and RES-04.
- scikit-learn 1.9 regression metrics and validation documentation.