Implementing unsupervised machine learning to gain a better understanding of the asphalt pavement conditions of Ontario provincial highways

Currently, the Ministry of Transportation of Ontario (MTO) obtains its pavement condition data via the Automated Road Analyzer (ARAN), an automated data collection vehicle and system, as part of its pavement management system activities. However, the pavement surface distress types that ARAN is able to discern are limited to cracking only. Without changing the formula used for the Pavement Condition Index, the pavement performance category could be misclassified. In this paper, instead of predicting target performance index values, the authors adopt unsupervised machine learning techniques, i.e., principal component analysis (PCA) and K-means clustering, to understand 2015 MTO asphalt pavement condition data that include 1,410 pavement sections. PCA is conducted to learn about the interrelationships among different key performance indices and employs the clustering method to categorize MTO provincial highways into three performance groups within each road functional class. In summary, this paper outlines an alternative approach to pavement condition assessment and could serve as a reference to facilitate decision-making for highway authorities.