Guiding fuzzy rule interpolation with information gains

Fuzzy rule interpolation enables fuzzy systems to perform inference with a sparse rule base. However, common approaches to fuzzy rule interpolation assume that rule antecedents are of equal significance while searching for rules to implement interpolation. As such, inaccurate or incorrect interpolated results may be produced. To help minimise the adverse impact of the equal significance assumption, this paper presents a novel approach for rule interpolation where information gain is utilised to evaluate the relative significance of rule antecedents in a given rule base. The approach is enabled by the introduction of an innovative reverse engineering technique that artificially creates training data from a given sparse rule base. The resulting method facilitates informed choice of most appropriate rules to compute interpolation. The work is implemented for scale and move transformation-based fuzzy rule interpolation, but the underlying idea can be extended to other rule interpolation methods. Comparative experimental evaluation demonstrates the efficacy of the proposed approach.