Fuzzy regression analysis using trapezoidal fuzzy numbers

When it is question of prediction, no deterministic model can be totally efficient, especially when the output to be estimated is dependent imprecisely on many fluctuant variables measuring human behavior (cognitions, choices, consumption, etc.). Regressions based on fuzzy logic which combine statistics and expert’s attitudes can be used to improve the estimation of such outputs. Those regressions are based on fuzzy logic which tries to traduce the human perception. In the literature, fuzzy linear regression has been developed following Tanaka’s model (the pioneer of such models) but the majority of the works make use of triangular fuzzy numbers and symmetric ones. In this paper, an extension to these regressions using trapezoidal fuzzy numbers is displayed. This suggested method is intending to use optimally the available data and gives the decider the opportunity to intervene and to use his experience in order to improve the quality of predictions.