Hybrid Neuro-Particle Swarm Optimization Model for Predicting Depression in Asphalt Pavements

Roads are critical to the economic development of a country. Pavement distresses affects the serviceability of the roads and road maintenance significantly disrupts the traffic flow and consequently the economy in the surrounding area. The application of machine learning techniques coincides with the shift of several industries to Industry 4.0. The objective of this study is to forecast the depression % occurrence in an asphalt pavement using an artificial neural network (ANN)-particle swarm optimization (PSO) algorithm. The network was developed using the temperature, precipitation, pavement age, and average annual daily traffic (AADT) as the input parameters (IP). The governing model developed using the ANN-PSO algorithm has an architecture of 4-9-1 (input-hidden-output). The governing model has the highest R and lowest Mean Squared Error (MSE). The mean absolute percentage error (MAPE) of the governing model is 6.38%. Using the connection weights (CW) of the governing model, the variable significance of the IP was obtained utilizing the Garson's algorithm (GA) and the AADT is the most influential parameter to the depression % occurrence. Moreover, the governing ANN-PSO model was compared to other prediction modeling methods including ensemble of trees, linear regression, regression trees, and support vector machine (SVM) and it was seen that the ANN-PSO is the superior model among the methods observed.