Interpretable fuzzy decision stability modeling for decision engineering under measurement and benchmark uncertainty

Abstract Fuzzy decisions may vary with measurement errors and admissible benchmark choices, including thresholds, weights, membership shapes, and aggregation rules. Existing fuzzy uncertainty representations and post-hoc sensitivity analyses do not generally certify decision invariance under their joint variation. This study proposes Benchmark-Multiverse Robustness-Certified Fuzzy Learning (BM-RCFL), integrating benchmark-conditioned fuzzy inference, conservative class-score bounds, object-level robustness certification, fragility diagnostics, and safe benchmark pruning. The evaluated implementation uses a fixed prototype-based reference model rather than end-to-end certificate-aware training. On synthetic data, BM-RCFL achieved an accuracy of $$1.0000$$ , a mean certified radius of $$0.3538$$ , and a certified decision rate of $$99.29\%$$ at $$\rho _i \ge 0.20$$ . On a Statistics Indonesia welfare panel with constructed reference classes, reference agreement was $$0.6404$$ , the mean certified radius was $$0.2028$$ , and $$57.02\%$$ of cases met the same threshold. No sampled perturbations within the certified regions changed decisions, providing an empirical implementation-consistency check rather than a mathematical proof. Full-enumeration verification using the same conservative bounds produced identical certificates, while benchmark pruning accelerated certification by factors of $$1.93$$ and $$3.14$$ on the respective datasets. Adversarial search identified decision reversals beyond the certified regions, and sensitivity analyses demonstrated dependence on benchmark widths and perturbation-metric scaling. BM-RCFL therefore provides an interpretable, assumption-dependent decision-stability assessment rather than a guarantee of decision correctness.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-12
DOI
https://doi.org/10.1038/s41598-026-76938-3
Primary Topic
Explainable Artificial Intelligence (XAI)
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Interpretable fuzzy decision stability modeling for decision engineering under measurement and benchmark uncertainty

Hiba Alsmadi, Koredianto Usman, Ayesha Iftikhar, Asma A. Alhashmi et al.
Scientific Reports
Explainable Artificial Intelligence (XAI)
article

Interpretable fuzzy decision stability modeling for decision engineering under measurement and benchmark uncertainty

Hiba Alsmadi, Koredianto Usman, Ayesha Iftikhar, Asma A. Alhashmi, Amirul Aizad Ahmad Fuad, Abdulbasit A. Darem, Asrul Sani, Rustam, Chiara Bordin
article en

Abstract

Abstract Fuzzy decisions may vary with measurement errors and admissible benchmark choices, including thresholds, weights, membership shapes, and aggregation rules. Existing fuzzy uncertainty representations and post-hoc sensitivity analyses do not generally certify decision invariance under their joint variation. This study proposes Benchmark-Multiverse Robustness-Certified Fuzzy Learning (BM-RCFL), integrating benchmark-conditioned fuzzy inference, conservative class-score bounds, object-level robustness certification, fragility diagnostics, and safe benchmark pruning. The evaluated implementation uses a fixed prototype-based reference model rather than end-to-end certificate-aware training. On synthetic data, BM-RCFL achieved an accuracy of $$1.0000$$ , a mean certified radius of $$0.3538$$ , and a certified decision rate of $$99.29\%$$ at $$\rho _i \ge 0.20$$ . On a Statistics Indonesia welfare panel with constructed reference classes, reference agreement was $$0.6404$$ , the mean certified radius was $$0.2028$$ , and $$57.02\%$$ of cases met the same threshold. No sampled perturbations within the certified regions changed decisions, providing an empirical implementation-consistency check rather than a mathematical proof. Full-enumeration verification using the same conservative bounds produced identical certificates, while benchmark pruning accelerated certification by factors of $$1.93$$ and $$3.14$$ on the respective datasets. Adversarial search identified decision reversals beyond the certified regions, and sensitivity analyses demonstrated dependence on benchmark widths and perturbation-metric scaling. BM-RCFL therefore provides an interpretable, assumption-dependent decision-stability assessment rather than a guarantee of decision correctness.

Scientific Reports
Universiti Malaysia Pahang Al-Sultan Abdullah (MY), Northern Border University (SA), University of Lahore (PK), Universitas Halu Oleo (ID), Al-Qalam University for Humanities and Applied Sciences (YE), UiT The Arctic University of Norway (NO), Telkom University (ID), Teesside University (GB)
Universitas Telkom
Explainable Artificial Intelligence (XAI)
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon