Bayesian model averaging

E835242

Bayesian method ensemble method model combination method statistical technique

Bayesian model averaging is a statistical technique that combines predictions from multiple models by weighting them according to their posterior probabilities to account for model uncertainty.

Jump to: Statements Referenced by

Statements (50)

Predicate	Object
instanceOf	Bayesian method ⓘ ensemble method ⓘ model combination method ⓘ statistical technique ⓘ
addresses	model selection uncertainty ⓘ
appliesTo	classification models ⓘ generalized linear models ⓘ hierarchical models ⓘ regression models ⓘ time series models ⓘ
basedOn	Bayes theorem NERFINISHED ⓘ
canUseApproximation	Bayesian information criterion NERFINISHED ⓘ Laplace approximation NERFINISHED ⓘ Markov chain Monte Carlo NERFINISHED ⓘ Occam’s window NERFINISHED ⓘ reversible jump MCMC ⓘ
combines	multiple candidate models ⓘ
contrastsWith	frequentist model averaging ⓘ single-model selection ⓘ
hasAdvantage	can improve out-of-sample prediction ⓘ propagates model uncertainty into parameter estimates ⓘ reduces overconfidence from conditioning on a single model ⓘ
hasChallenge	computational complexity for large model spaces ⓘ sensitivity to prior choices ⓘ specification of model priors ⓘ
hasComponent	model space ⓘ model-averaged predictions ⓘ model-specific parameters ⓘ posterior over models ⓘ prior over models ⓘ
hasPurpose	account for model uncertainty ⓘ improve predictive performance ⓘ incorporate model uncertainty into inference ⓘ
relatedTo	Bayesian model comparison ⓘ Bayesian model selection NERFINISHED ⓘ ensemble learning ⓘ stacking of predictive distributions ⓘ
typicalAssumption	one of the candidate models is close to the data-generating process ⓘ set of candidate models is specified ⓘ
usedIn	biostatistics ⓘ econometrics ⓘ environmental statistics ⓘ epidemiology ⓘ forecasting ⓘ machine learning ⓘ
uses	model likelihoods ⓘ posterior model probabilities ⓘ predictive distributions ⓘ prior model probabilities ⓘ
weightsBy	posterior probabilities of models ⓘ

Referenced by (1)

Full triples — surface form annotated when it differs from this entity's canonical label.

Bayesian linear regression → supports → Bayesian model averaging ⓘ