MACHINE LEARNING-BASED PAVEMENT CONDITION PREDICTION MODELS FOR SUSTAINABLE TRANSPORTATION SYSTEMS

Abstract

This study conducts a quantitative investigation into machine learning–based pavement condition prediction models, emphasizing their role in advancing sustainable transportation infrastructure management. Using a cross-sectional research design, the analysis integrates statistical modeling, computational learning, and environmental assessment to quantify how traffic, climatic, and material variables jointly influence pavement deterioration. Data were sourced from national pavement databases, automated condition surveys, and sensor-based monitoring systems, covering multiple climatic zones and roadway classifications. The methodological framework comprised two phases—model development and validation—featuring data preprocessing, feature optimization, and predictive modeling using algorithms such as random forest, gradient boosting, and artificial neural networks. Model evaluation employed statistical indicators including the coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE), supported by cross-validation and residual diagnostics to ensure robustness and reproducibility. The findings revealed that temperature anomalies and traffic load intensity were the most influential predictors, explaining nearly 87% of the variance in pavement condition indices. AI-enhanced regression models demonstrated up to an 8% improvement in predictive accuracy compared to traditional statistical methods, confirming their superior capacity to capture nonlinear deterioration patterns. Correlation analyses identified strong positive associations between thermal stress, axle repetitions, and distress frequency, while material thickness and binder resilience exhibited protective effects against deterioration. Reliability and validity testing produced high Cronbach’s alpha and composite reliability values, affirming measurement stability and structural coherence. Overall, the results validate that integrating machine learning with quantitative modeling enhances predictive precision, supports data-driven maintenance prioritization, and contributes to sustainable infrastructure planning by optimizing resource use, extending pavement lifespan, and minimizing environmental impact.