Predicting Pavement Structural Condition Using Machine Learning Methods

State departments of transportation recognize the need to incorporate pavement structural condition in their performance models and/or decision processes used select candidate projects for preservation, rehabilitation, or reconstruction at network level. However, data are costly obtain. To this end, paper develops and evaluates effectiveness two machine learning methods, Random Forest (RF) eXtreme Gradient Boosting (XGBoost), predicting a flexible pavement’s condition. The aim is be able predict whether section’s poor not based on Annual Average Daily Traffic (AADT), truck percentage, speed limit. considered if Surface Curvature Index (SCI12) above 3.3. developed using 950 miles Speed Deflectometer (TSD) collected along 8 primary routes South Carolina. was compared with that logistic regression model. When trained applied test data, prediction results indicated XGBoost RF outperform model by 12% 8%, respectively. outperformed 4%. With found best among three evaluated, its examined other threshold values; accuracy robust across different scenarios. AADT percentages significant factors whereas limit has no effect