Fuzzy higher type information granules from an uncertainty measurement

This paper proposes a new method for the formation of fuzzy higher type granular models. This is accomplished by directly discovering uncertainty from a sample of numerical information. In this case the coefficient of variation is proposed as a heuristic for measuring uncertainty, where a direct relation between this measure and the footprint of uncertainty of an interval type-2 fuzzy membership function is given. Followed by a steepest descent algorithm which is used to calculate interval Sugeno consequents for a fuzzy inference system. Two synthetic and two real datasets are used, measuring the performance by evaluating the output interval coverage of the formed granular models as well as the coefficient of determination which assesses modeling performance.