Modelling framwork
Suppose we have some random variable we want to predict
(over a space ) and some set of features or predictors in to make predictions of this are sampled from a random variable . You assume there is some relationship between and given by Where is some Irreducible error within our system and . Then you normally pick a modelling paradigm, like Linear regression, which relates to a subset of functions from
to , with an objective function . Then pick the best with respect to the objective function . Therefore you are making the prediction