A general fuzzy linear model for fuzzy regression analysis was first formulated. Based on both this general model and a fuzzy difference ranking method [2,4], approaches for fuzzy least squares regression and fuzzy least absolute value deviations regression were proposed. The former approach resulted in a nonlinear programming problem while the latter resulted linear. Numerical examples were solved by using the absolute deviations approach to illustrate the problems of conflicting trends and ways to at least partially overcoming these problems. Furthermore, these examples showed that the absolute deviations formulation forms an effective computational tool.
Relation:
IEEE International Conference on Fuzzy Systems Volume 2, 1994, Pages 1365-1369
