Smartphone-Based Cost-Effective Pavement Performance Model Development Using a Machine Learning Technique with Limited Data

Abstract

Road networks play a significant role in each country’s economy, especially in countries such as Afghanistan, which is strategically located in the international transit path from Europe to East Asia. In such a country, pavement performance models are fundamental for the pavement maintenance planning that provides high-quality infrastructure for transporting goods and travelers. However, due to the lack of a budget for pavement monitoring and maintenance in Afghanistan, transportation networks and pavement condition data have not been widely acquired for the development of a pavement performance model. The main aim of this study is to use a machine learning technique to, for the first time, develop a pavement performance model for Afghanistan that uses simple, cost-effective, and fairly accurate data—collected via smartphones—and that is based on a case study of over 550 km of Afghanistan’s highways. First, the current condition of Afghanistan’s road network is investigated using a smartphone. Then, collected data are prepared and analyzed so as to estimate the pavement condition index (PCI). Finally, a pavement performance model for PCI is developed using pavement age with an adequate coefficient of determination of 0.70 and successfully validated. It is concluded that the proposed approach is efficient and effective when developing a performance model in other developing countries encountering such data and budget limitations.