Similarity Function-Assisted Dynamic Fuzzy Rule Interpolation: An Improved Approach

Fuzzy rule interpolation (FRI) enables fuzzy inference systems to derive consequences when the observations are not covered by a system's sparse rule base. However, in conventional FRI systems, interpolated fuzzy rules are discarded once outcomes are obtained, despite they may contain valuable information about the problem space. Recent work has revealed the potential to improve the knowledge scope covered by a sparse rule base by reusing the interpolated rules. Particularly, dynamic fuzzy interpolation based on rule assessment (RAD-FRI) can reinforce the sparse rule base by adding high-quality interpolated rules into the rule base. It works via exploiting the similarity between interpolated rules and those given in the spare rule base. This paper further develops such research by improving the similarity function employed within RAD-FRI, through considering the location of the rules in a sparse rule base while filtering the interpolated rules that are not used in the subsequent inference processes. Comparative experimental outcomes on benchmark datasets demonstrate that the performance of the resulting FRI systems can be dynamically strengthened, with improved inference accuracy over that attainable by the popular existing FRI systems.