Modeling Asphalt Pavement Frictional Properties using Different Machine Learning Algorithms

Abstract

The objective of this work is to use some machine learning algorithms and test its efficiency in developing models to predict Locked Wheel Skid Trailer (LWST) values from Dynamic Friction Tester (DFT) and Circular Texture Meter (CTM) measurements conducted on asphalt pavement surfaces. For this prediction, three models were developed using DFT measurements at different speeds starting from 20km/h (12.5 mph) up to 64 km/h (40 mph) and then same DFT measurements as combination with Mean Profile Depth (MPD) and the last model used the International Friction Index (IFI) parameters (F60 and SP). The machine learning techniques includes two supervised learning algorithms: the Multi-Layer Perceptron (MLP) type of Artificial Neural Networks (ANN) and M5P tree model. In addition to one lazy algorithm called the K Nearest Neighbor (KNN) or Instance-Based Learner (IBL). The results showed that MLP models are the best in terms of the correlation coefficient that resulted in 81% prediction power using DFT parameters. Additionally, it was shown that the result of tree models was close to ANN but with much simpler regression. However, KNN models were recommended for LWST prediction of similar data characteristics and it is expected that this algorithm will be more efficient as the training data set becomes larger.