Harnessing Machine Learning Models for Pavement Texture Clustering

Pavement texture is a critical element affecting road safety and ride quality. It is affected by traffic volume, climate conditions, aggregate properties, and asphalt volumetric properties. This research aims to study the effect of different parameters on pavement texture using statistical and machine learning models. Pavement profile data and multiple variables affecting texture were collected from 192 SPS sections from the Long-Term Pavement Performance (LTPP) database. After data collection, pavement texture data were obtained from the pavement profile using ProVAL software and Python. Thereafter, the pavement texture was clustered into four diverse groups using the Gaussian Mixture Model (GMM), and the research determined cluster-specific profiles by applying centroid-based optimization techniques. Finally, an ordered logistic regression model and different machine learning models using K-nearest neighbor, random forest, extra trees, extreme gradient boosting, cat boosting, neural network, and weighted ensemble algorithm were developed to explore the parameters affecting the texture at diverse levels. The important parameters obtained from the statistical model were International Roughness Index (IRI), Annual Average Daily Truck Traffic (AADTT), temperature, and untreated subgrade, and from machine learning models were precipitation, IRI, AADTT, and 18-kips ESAL. Overall, this study significantly contributed to advancing the understanding and application of diverse impactful factors for pavement surface characteristics, pavement safety, and ride quality.