Prediction of Resilient Modulus of Subgrade Soils Using Machine Learning Techniques

Abstract

The accurate prediction of resilient modulus (MR) in subgrade soils remains a critical challenge in pavement design, as traditional methods like the California bearing ratio (CBR) fail to capture the dynamic nature of traffic loads. This study leverages machine learning techniques to predict MR using an extensive dataset from the long-term pavement performance (LTPP) program, comprising over 1000 samples from 60 countries. The methodology involved rigorous data preprocessing, including merging multiple sheets, calculating median MR values, handling missing data through statistical methods, estimating CBR values, and removing outliers. Four machine learning models—multiple linear regression (MLR), random forest (RF), K-nearest neighbors (KNN), and artificial neural network (ANN)—were developed and evaluated. The ANN model emerged as the superior predictor, achieving an exceptional R2 value of 0.95 with RMSE and MAE values of 1.995 and 3.98, respectively. The MLR and RF models also performed well, with R2 values of 0.82 and 0.81, while the KNN model achieved an R2 of 0.75. The strong performance across a diverse international dataset demonstrates this model’s potential for widespread application in various geological settings. The results particularly highlight the effectiveness of ANN in capturing complex, nonlinear relationships in soil parameters, suggesting that machine learning techniques, especially deep learning methods, could serve as valuable tools in geotechnical engineering applications, complementing traditional practices in pavement design and analysis. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2026.