Improving the Robustness of Data-Driven Pipelines via the Adaptive Neuro-Fuzzy Inference System: An Industrial Use Case
Year: 2025
Type: conference-paper
Abstract: Modern manufacturing is faced with challenges such as rapidly changing operations, real-time decision making, and uncertain data. As a result, the demand for accurate predictive analytics is becoming increasingly critical. This study presents an integrated Adaptive Neuro-Fuzzy Inference System (ANFIS), a data-driven pipeline that combines fuzzy rule systems with machine learning to improve defect prediction, enabling informed decision-making even in the presence of uncertainty and noisy data. The approach combines fuzzy logic to manage imprecise data, while machine learning enables adaptive learning from data. Applied to a real-world manufacturing case study, the proposed ANFIS model achieved an accuracy of $\mathbf{9 2 \%}$, precision of $\mathbf{8 7 \%}$, and F1-score of 84% in defect prediction, outperforming standalone fuzzy systems ($69 \%, 55 \%$, and 55%) and artificial neural networks ($\mathbf{8 8 \%, ~ 8 2 \%}$, and 79%). This approach contributes with an additional viable technique to an automated data-pipeline, proven to improve model accuracy and adaptability under certain criteria, supporting sustainable and intelligent decision making in line with Industry $\mathbf{5. 0}$ principles.
Authors: Aaqib Lone, Armir Bujari
Institution: University of Bologna
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4413319493