Application of machine learning techniques for predicting municipal pavement condition index

Machine learning (ML) is becoming a powerful tool for forecasting pavement performance and enhancing the efficiency of maintenance and resource planning. This study investigates the application of machine learning, using the random forest (RF) algorithm, to predict pavement condition index (PCI) over time. Manually assessed data from 2014 and 2018 were used to forecast the PCI for 2022, allowing the model to capture deterioration patterns across residential and non-residential streets in Skellefteå Municipality, Sweden. Model performance was evaluated using several performance metrics to assess flexibility and generalization across a diverse set of input variables, including pavement age (A), several distress mechanisms (D), previous PCI rating (S), and traffic data (T). For non-residential streets, models were further grouped by surface treatment type to evaluate RF’s predictive accuracy under varying maintenance conditions. Among residential street models, the RF model using A+D+S achieved the best results, with an R² of 0.78 and a correlation coefficient of 0.89 using the combined 2014 and 2018 dataset. For non-residential streets, incorporating traffic variables (T) further improved accuracy, with the RF(A+D+S+T) model reaching an R² of 0.79 and a correlation coefficient (ρ) of 0.88. While performance varied slightly across treatment types, the PCI rating assessed in 2018 (Status2018) and the weighted distress value (WDV) emerged as the strongest predictors. Furthermore, using the combined dataset (2014 and 2018) did not significantly improve prediction accuracy over using 2018 data alone. The findings demonstrate that RF models can effectively predict PCI and enhance decision-making in municipal pavement management.