Prediction of Pavement Condition Index using Artificial Neural Networks

Abstract

Modern pavement management systems aim to maintain performance and prioritize maintenance, reducing costs, and optimizing resources. Pavement Condition Index (PCI) is one of the most accurate tools to assess the structural condition of pavement, as it relies on evaluating various pavement distresses and converting them into a numerical value describes the pavement condition. This study aims to build a machine learning model to predict the pavement condition index using artificial neural networks (ANN). Three ANN models with different number of layers and nodes were designed for comparison. All models had one input layer consisting of seven types of pavement distresses, and one output layer consisting of the PCI value. Three models had one hidden layer with 4, 8, and 16 nodes. The data was normalized with a single weight scale by converting the descriptive data into numerical values and then calculating the weights for each type of distress according to its severity level. The results revealed that models had a coefficient of determination exceeding 0.95, which indicated the capability of artificial neural networks to model complex nonlinear relationships. Comparing the three models, the (7-8-1) architecture demonstrated the highest level of reliability with R2 achieving 0.98 and a standard error of less than 4%. In addition, K-fold cross-validation approach with 5-folds was conducted to validate the model. The average results of folds showed R2=0.95 and RMSE=5.72±0.49 which confirm the model's reliability in calculating PCI and assisting pavement M/R decision makers.