Unbiased use of data for input selection in fuzzy modelling
Jyh‐Shing Roger Jang · 2002
This paper describes an efficient method that computes the leave-one-out error for a linear model. The proposed method is able to find an unbiased performance index of a modeling approach involving the use of the least-squares method. The obtained performance index can then be used for model structure determination. A simple example of polynomial fitting is used to show the feasibility of the method. An advanced example of dynamical system identification via adaptive neuro-fuzzy inference system modeling is used to demonstrate the proposed method in practice for input selection in neuro-fuzzy modeling.