Abstract

This project explores the development and optimization of predictive models for the resilient modulus (MR) of subgrade soil using advanced machine learning techniques. Comprehensive data from INDOT spanning several years was analyzed to enhance the accuracy of MR predictions. The study not only refined the modeling approach through statistical methods and validation but also identified crucial soil properties that significantly impact MR values. Recommendations for future data collection were made to further improve the models. The developed models and these recommendations will be used to guide INDOT in making informed decisions for pavement design and maintenance, which will ultimately lead to more efficient and cost-effective engineering practices.

Keywords

clustering, machine learning, model validation, random forest, resilient modulus, XGBoost, regression, repeated load triaxial test, resilient modulus

Report Number

FHWA/IN/JTRP-2024/27

SPR Number

4714

Performing Organization

Joint Transportation Research Program

Publisher Place

West Lafayette, Indiana

Date of Version

2024

DOI

10.5703/1288284317768

12_12_23.ipynb (4377 kB)
MR Main Code

XGBoost for k values.ipynb (764 kB)
Calibrated_constitutive_model_blackbox

k1.xlsx (776 kB)
Closedform_k1_universal

K2.xlsx (685 kB)
Closedform_k2_universal

k3.xlsx (691 kB)
Closedform_k3_universal

INDOT_data_final.csv (175 kB)
Cleaned Dataset

INDOT MR explicit models.xlsx (10 kB)
Linear and Polynomial Models Excel File

SPR-4714 Technical Summary.pdf (332 kB)
Technical Summary

Share

COinS