Нелинейная связь не может быть представлена только straight-line combination of variables. Нелинейные models могут захватывать curved, interaction-based или more complex patterns, которые simple linear models cannot express.
A relationship between two or more variables that can't be represented solely through addition and multiplication. A linear relationship can be represented as a line; a nonlinear relationship can't be represented as a line. For example, consider two models that each relate a single feature to a single label. The model on the left is linear and the model on the right is nonlinear: !Two plots. One plot is a line, so this is a linear relationship. The other plot is a curve, so this is a nonlinear relationship. See Neural networks: Nodes and hidden layers in Machine Learning Crash Course to experiment with different kinds of nonlinear functions.