Determining Linguistic Models with Constrained Fuzzy Regression

Abstract

Principled, automated procedures for assigning optimal linguistic representations to the output (or input) data space of a fuzzy control universe do not exist at the present time. This paper suggests that locating and verifying such assignments via constrained fizzy linear regression can provide the necessary foundation for their (automated) optimization. The procedure described here is related to methods for the resolution of (so-called) illposed problems and consists essentially of repeated iterations of fizzy linear regression and (crisp) linear programming operations over linguistic representations of Jicuy numbers which "cover" the raw data.. lbis paper demonstrates that a global approach to the problem of Jinding the optimal linguistic representations for raw cfuzzy control) data may prove simpler, more eflective, and more readily machine-learnable than the local approaches used heretofore.