Automatic pavement damage predictions using various machine learning algorithms: Evaluation and comparison

In recent years, different technologies based on Machine Learning (ML) algorithms have been developed to evaluate road distress quality and help with pavement maintenance, traffic safety, and road rehabilitation. However, the literature lacks a comparative study of various ML algorithms to detect pavement damage. This study aims to fill that gap by comparing various regression algorithms including Linear Regression (LR), Support Vector Regressor (SVR), Random Forest Regressor (RFR), k-nearest neighbors, Gradient Boosting Regressor (GBR), Light Gradient Boosted Machine (LGBM), stacking regressor, and Decision Tree Regressor (DTR). To evaluate the accuracy of the models, different performance metrics, namely coefficient of determination (R2), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE) are used. Experiments are conducted to validate the models' performance using a dataset sponsored by the Michigan Department of Transportation and executed at Michigan State University for the Michigan road pavement damage. Obtained results indicate that the GBR modeling algorithm performs better with the most accurate asphalt pavement damage model of R2 = 99%, MAE = 0.02, and RMSE = 0.03. Thus, the use of GBR modeling might be an easy, robust, and efficient method for automated pavement damage prediction.

Automated summary

In recent years, different technologies based on Machine Learning algorithms have been developed to evaluate road distress quality and help with pavement maintenance and rehabilitation. This study compares various regression algorithms and finds that the GBR modeling algorithm performs the best in predicting pavement damage.