A decision boundary is a hypersurface that partitions the underlying vector space into two sets, one for each class in a binary classification problem. It is determined by the model and represents the threshold at which the model switches from predicting one class to another.