Forward selection begins with no variables in the model. Variables are added one by one based on a specified criterion, usually the p-value or AIC. At each step, the variable that improves the model the most is added. The process continues until no significant improvement is observed.
