Data Cleaning Framework for Pavement Maintenance and Rehabilitation Decision-Making in Pavement Management System Based on Artificial Neural Networks

Abstract

The quality of data in a pavement management system (PMS) has a direct impact on pavement maintenance and rehabilitation (M&R) decisions and management. However, pavement performance data and M&R action data often suffer from problems such as omissions and anomalies. To solve these problems, this study proposes a data cleaning framework based on artificial neural networks (ANNs) that can clean pavement performance data and M&R action data simultaneously. First, data are classified and labeled by the framework using the percentile method and considering the nature of the PMS data itself. Then two ANNs are established, one to clean data from anomalous or omitted pavement performance, and another to fill in data from omitted M&R actions. Applying the framework to the PMS in Shanxi Province, China, the following conclusions can be drawn. In terms of filling in the omitted M&R action, ANN calculated less average loss and improved the average prediction accuracy by 7.88% compared to the logistic regression model, proving the superiority of ANN in filling in the omitted M&R action data. Compared to the framework of filling in the omitted M&R action data by ANN without cleaning the pavement performance data, the proposed framework resulted in less average loss values and 5.71% improvement in average accuracy, demonstrating the need for cleaning both types of data simultaneously. The framework can provide a higher-quality data set for pavement M&R decisions and management.