Generating Uncertain Fuzzy Logic Rules from Surveys: Capturing Subjective Relationships between Variables from Human Experts

One of the biggest challenges in the design of Fuzzy Logic Systems (FLSs) is the construction of their rule base. While fuzzy sets capture aspects of a system's variables and associates them with linguistic labels, it is the rules which capture the logical relationships of these labels and underlying fuzzy sets. Further, while fuzzy systems are credited for dealing well with uncertainty in system inputs and outputs, comparatively little research has focused on the capture of uncertainty in their actual inference rules. This paper focusses on the challenge of capturing the knowledge of multiple human experts on the relationships of linguistic labels in a given problem domain. Specifically, it proposes a novel survey-centric methodology which enables the capture of individual, subjective input from domain (not fuzzy logic) experts with minimal prior training and provides mechanisms to aggregate the resulting survey-data into a working and interpretable fuzzy system. The rule base of the resulting system incorporates weights to capture intra- and inter-expert uncertainty during rule specification. The paper follows a practical style to facilitate reproduction of the proposed methodology by peers. Results and initial evaluation based on real world case studies in the context of environmental conservation in Western Australia are provided.