Fuzzy hybrid system based on ensemble learning for classification in intelligent manufacturing processes

Abstract

Abstract The integration of Machine Learning (ML) into advanced manufacturing is a key enabler of Industry 4.0, supporting intelligent quality control and early fault detection. Despite recent advances, challenges remain regarding the interpretability and adaptability of multiclass classification models operating under uncertainty in industrial environments. This paper proposes a hybrid approach that combines ensemble ML models with fuzzy logic, integrating the outputs of probabilistic classifiers through a rule-based inference system that incorporates prior distributions. The proposed framework provides a flexible and interpretable aggregation mechanism that supports both homogeneous and heterogeneous ensembles. A case study was conducted to classify tire performance during uniformity testing at a multinational manufacturing plant. Five supervised ML algorithms were evaluated, and explainable artificial intelligence (XAI) techniques were employed to identify the variables that most strongly influence classification performance. Experimental results showed that the proposed fuzzy-based system achieved predictive performance statistically equivalent to the corresponding soft-voting ensembles while significantly outperforming several individual classifiers. The best-performing configuration, combining the Random Forest (RF), Support Vector Machine (SVM), and Gradient Boosting Decision Tree (GBDT) classifiers, achieved a mean accuracy of 84.1%, suggesting that fuzzy aggregation may represent a promising and interpretable alternative for multiclass classification in intelligent manufacturing. These findings provide suggestive evidence supporting the potential application of fuzzy ensemble learning to quality control and decision-making processes, particularly in tire manufacturing.