Application of Artificial Neural Networks for Predicting the Structural Number of Flexible Pavements Based on Subgrade Soil Properties
Authors:
Asadullah Ziar 1*, Wahidullah Hazim 2
M.Sc. in Civil Engineering, Department of Civil Engineering, Ghazni Technical University, Afghanistan
1
Bachelor of Engineering, Department of Civil Engineering, Kabul University, Afghanistan
2
doi.org/10.52132/Ajrsp.e.2025.80.1
The structural number (SN) is a critical parameter in the AASHTO design method, representing the overall load-bearing capacity of flexible pavements. Traditional determination of SN requires resilient modulus (MR) and California Bearing Ratio (CBR) tests, which are both costly and time consuming. This study proposes an artificial neural network (ANN) model as an alternative approach for predicting SN using readily available subgrade soil properties and environmental factors. A dataset of 2,810 samples was compiled and preprocessed, with dry unit weight (γd), moisture content (w), weighted plasticity index (wPI), and number of freeze–thaw cycles (NFT) employed as model inputs. The ANN was developed in MATLAB using a feed-forward architecture with a single hidden layer of 10 neurons and trained with the Levenberg–Marquardt algorithm. Model performance was evaluated using mean squared error (MSE) and correlation coefficient (R). The results showed strong predictive capability, with R values of 0.954, 0.948, 0.942, and 0.951 for training, validation, testing, and overall datasets, respectively. Error histograms and regression plots confirmed the model’s robustness and generalization capacity. The proposed ANN framework provides a reliable and cost-effective tool for estimating SN, reducing dependence on expensive laboratory testing while supporting efficient and accurate pavement design.
Keywords:
Artificial neural networks (ANNs), flexible pavement design, structural number (SN), subgrade soil properties, machine learning in transportation engineering.
Academic Journal of Research and Scientific Publishing
