Machine Learning-Based Prediction of International Roughness Index for Road Maintenance

Abstract

Road roughness is a key indicator of pavement condition and significantly affects transportation safety, ride quality, vehicle operating costs, and maintenance planning. The International Roughness Index (IRI) is widely used by road administrations to assess pavement performance and support maintenance decision-making. IRI accurate prediction is important to optimise the maintenance strategies by upgrading long term sustainability of road infrastructure management. This thesis investigates the application of machine learning techniques for predicting IRI changes on the Swedish E4 highway, using data obtained from Trafikverket's Pavement Management System (PMS4) and the Road Weather Information System (VVIS). The study integrates pavement condition, traffic, maintenance, structural, and climatic variables to predict IRI changes based on data from approximately 130 km of the E4 corridor. Several predictive models were evaluated, including Random Forest and Ridge Regression. Following feature selection and correlation analysis, a final set of twelve explanatory variables was retained. Model performance was evaluated using K-fold cross validation together with commonly used regression performance indicator, namely the coefficient of determination (R²), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). The results demonstrate that the Random Forest model outperformed Ridge Regression, achieving an R² of 0.70, an RMSE of 0.13, and an MAE of 0.078. Previous IRI values, maintenance history, rut depth, and traffic loading were identified as the most influential predictors. Weather-related predictors likewise demonstrated a meaningful contribution to overall model accuracy, reinforcing the significance of incorporating environmental exposure factors within pavement degradation forecasting framework. The findings indicate that Random Forest provides a reliable framework for pavement roughness prediction and can support data-driven maintenance planning and long-term asset management within the Swedish road network.