A Tri-Criterion Framework for Interpretable, Adaptive, and Robust Fuzzy Decision-Making Under Uncertainty An IRA-FIS Approach with an Environmental Health Risk-Assessment Case Study

Abstract

Fuzzy decision-making provides a principled way to model uncertainty, imprecision, and subjectivity in
complex real-world systems, yet three properties that largely determine whether a fuzzy model is actually adopted
in practice — interpretability, adaptability, and robustness — are rarely evaluated jointly within a single,
reproducible procedure. This paper introduces IRA-FIS, a fuzzy inference framework that couples a Mamdani
type rule base with three explicit, quantifiable criteria: a rule-complexity interpretability index, a bounded online
adaptability mechanism for membership-function drift, and a robustness index derived from Monte Carlo
perturbation of inputs. We formalize each criterion mathematically, define a configurable composite IRA score for
comparing competing fuzzy designs, and demonstrate the framework on an environmental health risk-assessment
case study that classifies exposure risk from three uncertain stressors: pollutant concentration, exposure duration,
and population vulnerability. Sensitivity analysis shows that the resulting system maintains stable centroid
outputs under input noise of up to 20% while preserving a compact, expert-auditable rule base of fewer than thirty
rules. We situate the contribution relative to recent neuro-fuzzy, type-2 fuzzy, and explainable-AI hybrids reported
between 2023 and 2026, and discuss limitations and avenues for extending IRA-FIS toward type-2 fuzzy sets and
constrained data-driven rule learning. The framework is offered as a methodological contribution directly aligned
with this Collection’s call for algorithmic developments and practical applications of fuzzy logic in decision
making that improve interpretability, adaptability, and robustness.