Comparative pavement performance modelling using machine learning and probabilistic methods

Pavement management plays a crucial role in a country’s economic framework, as road authorities strive to extend the lifespan of their invaluable road assets. This study applies a systematic approach to develop and compare pavement performance models, with a particular focus on machine learning and probabilistic methods. The initial phase involves utilising a smartphone to assess the pavement condition of Afghanistan’s regional highway. Subsequently, the collected data are prepared and analyzed to determine the type, severity, and density of pavement distress, which are used to compute the Pavement Condition Index (PCI). The study proceeds to develop pavement performance models for PCI based on pavement criteria encompassing pavement age, pavement thickness, traffic load, and weather conditions, applying 14 regression-based machine learning and Markov Chain probabilistic methods. The Markov Chain outperformed the machine learning methods. It is concluded that the suggested method is practical and efficient for developing a performance model.