THE USE OF ANN AND MACHINE LEARNING ALGORITHMS TO PREDICT ROAD SURFACE DETERIORATION

Abstract

Despite advancements in the application of artificial intelligence for monitoring and predicting pavement conditions, current models are not extensively utilized due to their limited adaptability and inadequate consideration of environmental variables. This study focuses on developing enhanced models for predicting the Pavement Condition Index (PCI) using artificial neural networks and the backpropagation algorithm. The aim is to improve the accuracy of the predictions. The models were trained using a dataset of 1,614 samples collected during an experiment conducted on a motorway between Kostanai and Astana. The dataset included information on asphalt pavement thickness, subgrade, traffic loads, temperature, precipitation, and deflectometer data. The architecture model with the highest performance, labeled as 9-9-1, attained peak efficiency with a value of 0.0344 after 22 training iterations. The results demonstrated a high level of accuracy, as indicated by a multiple correlation coefficient (R²) of 0.954, a mean absolute error (MAE) of 0.125, and a root mean square error (RMSE) of 0.162. The developed models possess the capability to extrapolate information, adjust to variations, and accurately forecast the rate of roadway deterioration.