Automating expert-based road maintenance evaluation under uncertainty using fuzzy machine learning for data-driven decision support

Road maintenance prioritisation still relies heavily on expert judgment, despite advances in pavement inspection technologies. Although machine learning has been widely applied in infrastructure assessment, most studies focus on predicting individual distress indicators rather than automating expert-based prioritisation. This study develops an automated fuzzy machine learning framework to learn and replicate expert engineering judgement for road maintenance evaluation under uncertainty. Quantitative pavement indicators are transformed using Type-1 triangular membership functions, while qualitative engineering attributes are incorporated through categorical encoding. Six regression algorithms representing linear, neural, kernel-based, and ensemble learning paradigms are evaluated within the fuzzy framework. Using 10-fold cross-validation on 420 road sections, model performance was assessed using multiple metrics and a composite Reference Index, with Friedman and Wilcoxon signed-rank tests used for statistical validation Fuzzy–Linear Regression achieved the best generalisation performance (RI = 0.966; R2 = 0.884) and significantly outperformed the tree-based models while remaining comparable to other nonlinear learning algorithms. Multi-regional validation further confirmed its robustness and transferability. The proposed framework provides a practical, data-driven decision-support tool for automating expert-based road maintenance prioritisation under uncertainty.process