A software for predicting Pavement Condition Index (PCI) using machine learning for practical decision-making with an exclusion approach

Abstract

In Palestine and other resource-constrained settings, determining the Pavement Condition Index (PCI) requires exhaustive visual surveys of up to 19 distress types, which is a process that is both time-consuming and costly to obtain. Despite advances in PCI prediction (2023–2025), existing methods still depend on full-distress assessments, failing to reduce fieldwork burden. We present an open?source machine learning software that classifies pavement into PCI categories (Good, Satisfactory, Fair, Poor, Impassable) by systematically excluding low-utility distresses, reducing inspection effort by up to 40% while achieving an overall accuracy of 82%. The framework integrates features such as pavement age, layer thickness, right-of-way (ROW), average daily traffic (ADT), and heavy-duty vehicle percentage.