方法对比

	多元自适应回归样条 (MARS)	广义可加模型 (GAM)	梯度提升（Gradient Boosting）
领域	机器学习	机器学习	机器学习
方法族	Machine learning	Machine learning	Machine learning
起源年份	1991	1986	2001
提出者	Jerome H. Friedman	Trevor Hastie & Robert Tibshirani	Friedman, J. H.
类型	Adaptive piecewise-linear regression	Semi-parametric additive regression model	Ensemble (sequential boosting of decision trees)
开创性文献	Friedman, J. H. (1991). Multivariate adaptive regression splines. The Annals of Statistics, 19(1), 1–67.	Hastie, T., & Tibshirani, R. (1986). Generalized additive models. Statistical Science, 1(3), 297–310.	Friedman, J. H. (2001). Greedy Function Approximation: A Gradient Boosting Machine. Annals of Statistics, 29(5), 1189–1232.
别名	multivariate adaptive regression splines, earth algorithm, MARS regression, çok değişkenli uyarlamalı regresyon spline'ları	GAM, additive model, spline-based additive regression, Genelleştirilmiş toplamsal model	Gradient Boosting (GBM), GBM, gradient boosted trees, gradient boosting machine
相关	4	4	5
摘要	Multivariate adaptive regression splines, introduced by Jerome Friedman in 1991, is a flexible nonparametric regression method that automatically models nonlinearities and interactions by combining piecewise-linear 'hinge' functions. It builds the model in a forward stagewise pass that adds basis functions where they help most, then prunes back the overgrown model, yielding an interpretable additive-plus-interaction form that adapts its complexity to the data.	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.	Gradient Boosting is an ensemble learning method, formalised by Jerome H. Friedman in 2001, that combines a sequence of weak learners — typically shallow decision trees — so that each new tree is fitted to minimise the residual errors of the trees before it. It is the core algorithm behind popular implementations such as XGBoost, LightGBM and CatBoost.