A generalized fuzzy weighted least-squares regression

Abstract A fairly general fuzzy regression technique is proposed based on the least-squares approach. The main concept is to estimate the modal value and the spreads separately. In order to do this, the interactions between the modal value and the spreads are first analyzed in detail. The advantages of this new fuzzy weighted least-squares regression (FWLSR) approach are: (1) the estimation of both non-interactive and interactive fuzzy parameters can be performed by the same method, (2) the decision-makers' confidence in the gathered data and in the established model can be incorporated into the process, and (3) suspicious outliers (or fuzzy outliers), that is, data points that are obviously and suspiciously lying outside the usual range, can be treated and their effects can be reduced. A numerical example is provided to show that the proposed method can be an effective computational tool in fuzzy regression analysis.