What it is
Regression estimates a continuous quantity such as travel time, energy load, water demand, or pollutant concentration. A regression output is an estimate conditional on selected data and assumptions. It should be accompanied by error measures and checked for systematic differences across locations and operating periods.
How it works
- 01Choose the quantity, unit, geographic scale, and forecast or estimate horizon.
- 02Inspect missing data, measurement error, outliers, and relationships among predictors.
- 03Fit a suitable baseline and candidate models using training data.
- 04Evaluate on later or held-out locations; examine residuals and communicate uncertainty.
- 01
Choose the quantity, unit, geographic scale, and forecast or estimate horizon.
- 02
Inspect missing data, measurement error, outliers, and relationships among predictors.
- 03
Fit a suitable baseline and candidate models using training data.
- 04
Evaluate on later or held-out locations; examine residuals and communicate uncertainty.
Why it matters
Numeric estimates help staff plan capacity and compare scenarios. Errors have different costs: underestimating a peak load may be more serious than a small overestimate, and a citywide average can conceal poor performance in a particular area.
Typical Smart City applications
- Building electricity or cooling load estimation
- Travel-time and transit demand estimates
- Water consumption forecasting
- Infrastructure condition or exposure measurement
Example data inputs
Strengths
- Produces interpretable numeric estimates for planning
- Can compare a simple baseline with more complex approaches
- Residuals help reveal missing structure and systematic error
Limitations and risks
- Correlated predictors do not show that one factor caused another.
- Extrapolation beyond observed conditions can be unreliable.
- Aggregate fit can conceal errors that matter for a particular service area.
Illustrative workflow
Applied city example
A facilities team estimates next-day electricity use for public buildings using historical meter readings and weather. It compares results with a seasonal baseline and reports forecast ranges to planners rather than treating one estimate as a guaranteed load.
Related methods
- Supervised learning
- Time-series forecasting
- Clustering
- Optimization algorithms
Frequently asked questions
Is regression the same as forecasting?
Regression estimates numeric outcomes. Forecasting emphasizes their future values over time; forecasting often uses regression components.
What is a residual?
The difference between an observed value and a model estimate. Patterns in residuals can reveal systematic misspecification.
Should a regression output include a confidence interval?
When feasible, uncertainty ranges help decision-makers understand how much the estimate may vary.