An Interval Approach for Fuzzy Linear Regression with Imprecise Data

In this paper, a revisited approach for fuzzy regression linear model representation and identification is introduced. By adopting the commonly used principle of alpha-cuts, the fuzzy regression implementation is reduced to the handling of conventional intervals, for inputs, parameters and outputs. Using the Midpoint-Radius representation of intervals, the uncertainty attached to linear models becomes more interpretable. Actually, it is possible to determine the output uncertainty origin (model parameters and/or inputs). In this context, a possibilistic regression method is proposed to identify models of minimal global uncertainty, that is with respect to all possible inputs.