Predicting the behaviour of geosynthetic-reinforced soil abutments using machine learning
Loading...
Date
2025
Journal Title
Journal ISSN
Volume Title
Publisher
UMT. Lahore
Abstract
Geosynthetic-reinforced soil (GRS) abutments are commonly used in bridge
construction, retaining walls, and highway embankments because they are strong, flexible, and
cost-effective. One of the most important things to consider when using these structures is how
much they might settle, since too much movement can affect safety and how well they work.
Traditional models for predicting settlement often make too many simplifying assumptions
about how the soil interacts with the reinforcement, while more advanced methods like finite
element modeling are accurate but need a lot of time, skill, and computing power.
This study uses a machine learning approach called Gene Expression Programming
(GEP) to predict how much GRS abutments settle.
The researchers used a dataset with 354 experimental observations, where the goal was
to predict settlement, and the inputs included factors like surcharge pressure, soil friction angle,
reinforcement stiffness, vertical spacing, abutment shape, and the angle of the facing. The GEP
model performed very well, with a testing score of 0.91, an RMSE of 5.53, and an MAE of
3.09, which shows it agrees closely with the actual experimental results. Additional checks
using separate data confirmed the model's reliability.
A sensitivity analysis also showed which factors had the biggest impact on settlement,
giving useful guidance for future design improvements. This research shows that machine
learning, especially GEP, can be a fast and dependable alternative to older methods for
predicting settlement. It supports a more performance-based and data-driven approach in
geotechnical engineering.