Ensemble-based approach for automatic prediction of pullout resistance of geogrids in different soil types

Determination of the pullout resistance of geogrid, an essential parameter in MSE wall design, is time-consuming and expensive. The present study applies ensemble methods, namely, random forest, gradient boosting, extreme gradient boosting (XGB), and light gradient boosting to predict the pullout resistance factor (F∗) of geogrid. An extensive review resulting in a large pullout test dataset of 759 data points encompassing various influencing features such as normal stress, relative compaction, fines content, average particle size of fill material, embedment length, ultimate tensile strength, and longitudinal and transverse spacing of ribs of the geogrid, and pullout displacement rate is used to evaluate models. Results showed that the XGB (R² = 0.91 and RMSE = 0.18) outperformed the other ensemble approaches. Based on the feature importance analysis on the best-performing XGB model, normal stress, reinforcement embedment length, and relative compaction are found to be the most influencing parameters affecting F∗. A simplistic model to predict F∗ as a function of only these three influencing parameters is proposed considering the ensemble model. Furthermore, limited laboratory pullout experiments are performed to evaluate these models. The proposed machine learning models fitted very well with the laboratory F∗ values with an error within ±3 %.