Linear regression is the simplest and most widely-used model for supervised learning with continuous targets. It is typically the fastest type of model to train and yields a linear prediction function that is particularly easy to interpret and to use in scoring observations.
In addition to fitting a standard linear regression model, the Linear-AS node allows you to perform model selection using forward stepwise or best subsets methods. Predictor selection for forward stepwise modeling can be based on standard F statistics, Adjusted R2, or minimizing the corrected Akaike Information Criterion (AICc) or the average squared error (ASE) over a randomly chosen overfit prevention subset of the data (about 30% of the instances are randomly held out in model estimation in this case). The node can also attempt to automatically fit two-way interaction terms.