Intelligent pavement management: a machine learning framework for condition classification and maintenance prioritization

Efficient pavement maintenance requires reliable classification of pavement condition and clear prioritization of interventions. This study develops and benchmarks a multi-model machine learning framework for pavement condition classification in both urban and rural road networks, integrating distress indicators, road geometry, and motorized and non-motorized traffic distributions. Eight supervised learning models (Logistic Regression, Naïve Bayes, SVM, KNN, Decision Tree, Random Forest, XGBoost, and ResNet) were evaluated using k-fold cross-validation, class-wise F1-scores, and confusion matrices. Tree-based ensemble models, particularly Random Forest and XGBoost, consistently demonstrated stable and balanced predictive performance across Good, Fair, and Poor pavement classes. Model outputs were translated into a five-tier maintenance prioritization framework (Critical, High, Medium, Low, Monitor), enabling actionable scheduling for pavement management. Results confirm the effectiveness of ensemble models for decision-support in pavement management, and highlight the importance of context-specific modeling for urban and rural environments.