Detection of Unreported Treatments in Pavement Management System of Iowa DOT Using Machine Learning Classification Algorithm

Abstract

Treatment records are among the most frequently underreported data items in pavement management systems (PMSs), which negatively affects various PMS analysis tools, such as pavement performance and deterioration models. Disregarding unreported treatments may lead to inaccurate pavement age and condition estimates, resulting in erroneous and nonoptimal maintenance and rehabilitation decisions. Nevertheless, the unreported and frequently missing pavement treatment data has received limited attention. Hence, this paper contributes to the body of knowledge by introducing a methodology for detecting unreported treatment actions and their occurrence probabilities over pavement age using a machine learning classification algorithm. Logistic regression models were developed using historical pavement condition data and validated on two levels: (1) split validation; and (2) manual validation using video logs of the pavement condition before and after treatment application. The results show that the developed models can detect unreported pavement treatments with accuracy, precision, and F1 scores ranging from 89% to 96%, 82% to 91%, and 70% to 85%, respectively. The presented methodology and developed models will help highway agencies identify unreported and missing pavement treatments, contributing to more cost-effective maintenance and rehabilitation decisions.