A Subsethood Measure Based Approach for Handling Sparse Rule Bases Without Rule Interpolation

This paper introduces an analogical reasoning scheme that is grounded in lattice theory and uses a novel subsethood measure to handle sparse fuzzy rule bases by directly activating the most relevant rules. This methodology allows us to forego the use of rule interpolation. Based on this framework, an algorithmic strategy is proposed to address regression and time series prediction, constructing a particular fuzzy inference system for each training point and combining their outputs using a distance-weighted approach. Our paper includes several experimental results that indicate a competitive performance with fewer rules compared to interpolation-based methods. Future research will focus on optimizing parameter selection, improving computational efficiency, and enhancing robustness.