Generalized Additive Model (GAM)
Also known as: GAM, additive model, spline-based additive regression, Genelleştirilmiş toplamsal model
A generalized additive model, introduced by Trevor Hastie and Robert Tibshirani in 1986, extends the generalized linear model by replacing each linear term with a smooth, data-driven function of the predictor. This lets the model capture nonlinear relationships while preserving the additive, term-by-term interpretability of regression: each predictor contributes its own estimated curve, and the curves simply add up (on a link scale) to predict the response.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use a GAM when you expect nonlinear relationships between predictors and the response but still want an interpretable, inspectable model — one where you can plot and reason about each predictor's effect. It is well suited to regression and classification with continuous predictors in epidemiology, ecology, economics, and the environmental sciences, and to any setting where a black-box model's opacity is unacceptable. The core form assumes effects are additive (no automatic interactions), though specific interactions and tensor-product smooths can be added explicitly. When interactions are pervasive and interpretability is secondary, tree ensembles or neural networks may predict better; when relationships are genuinely linear, a GLM is simpler and equally good.
Strengths & limitations
- Captures nonlinear effects while remaining interpretable one predictor at a time.
- Smoothing penalties and automatic λ selection guard against overfitting.
- Extends to non-Gaussian responses through the link function (logistic, Poisson, etc.).
- Provides effective degrees of freedom and confidence bands for each smooth term.
- Assumes additivity by default, missing interactions unless they are added explicitly.
- Higher-dimensional smooths and many predictors become harder to fit and interpret.
- Smoothing-parameter and basis choices influence the fit and require care.
- Less predictive than flexible black-box models when complex interactions dominate.
Frequently asked
How is a GAM different from linear regression?
Linear regression assumes each predictor has a straight-line effect; a GAM replaces each linear term with a flexible smooth function, so effects can be curved. It keeps the additive structure, so you can still interpret each predictor's contribution separately, but now those contributions need not be linear.
Does a GAM model interactions?
Not by default — the standard GAM is additive, meaning each predictor contributes independently. Interactions can be included deliberately through bivariate or tensor-product smooths, but if you do not add them, the model assumes no interaction. This is the main structural assumption to keep in mind.
How is overfitting controlled?
Each smooth carries a roughness penalty governed by a smoothing parameter. Larger penalties force simpler (straighter) curves. The smoothing parameters are typically chosen automatically by cross-validation or restricted maximum likelihood, balancing fit against smoothness so the curves are only as wiggly as the data support.
Sources
- Hastie, T., & Tibshirani, R. (1986). Generalized additive models. Statistical Science, 1(3), 297–310. DOI: 10.1214/ss/1177013604 ↗
- Hastie, T. J., & Tibshirani, R. J. (1990). Generalized Additive Models. Chapman & Hall/CRC. ISBN: 978-0-412-34390-2
How to cite this page
ScholarGate. (2026, June 2). Generalized Additive Model (GAM). ScholarGate. https://scholargate.app/en/machine-learning/generalized-additive-model
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- LOESSMachine learning↔ compare
- Multiple Linear RegressionStatistics↔ compare
- Polynomial RegressionStatistics↔ compare
- Regression SplinesMachine learning↔ compare