Definition
Regression models map inputs to real-valued outputs. Linear regression is the classic starting point; tree ensembles and neural nets handle nonlinear relationships.
Evaluation uses MAE, MSE, RMSE, or R² rather than classification accuracy.
In simple terms
Classification asks "what kind?"; regression asks "how much?" — like estimating a house price instead of labeling it luxury or budget.
Where you see it
- Forecasting demand or energy usage.
- Predicting reading time or model latency.
- Estimating property values from features.
How it works
1.Collect numeric targets
Each example has a continuous label.
2.Fit a model
Minimize error between predicted and true values.
3.Evaluate residuals
Check how wrong predictions are on held-out data.
Why it matters
- Many business and scientific questions need numbers, not categories — regression is the tool.
Often confused
Regression only means linear regression.
Any model that predicts continuous values is regression — including deep nets.