Random Forest is a non-linear model because it does not assume a linear relationship between the target and predictor variables. The non-linearity of Random Forest, or even a single decision tree, can be illustrated by the box-like decision boundary that results from the training of the model. For instance, it is possible for the same feature to be used in different splitting criteria within different levels of a decision tree. As a result, the algorithm is able to learn complex relationships present within the data structure without being subject to the constraints and assumptions of a linear model.