This occurs when the model used to describe a set of data is unnecessarily complicated. For example, if, with x taking the values (1, 10, 40, 60), the corresponding values of y are found to be (2, 20, 80.1, 120), then the model y=2x should suffice. Anything more complex would probably be described as overfitting. The term is used in machine learning to describe a complex model that fits known data well but is less successful than a simpler model at fitting the values of a subsequent data set.