Prediction of Design Parameters for Retaining Walls with Gaussian Process Regression

Creative Commons License

Çalık Ü.

VII. International Applied Statistics Congress, İstanbul, Turkey, 11 - 13 May 2026, vol.1, no.1388, pp.470-485, (Full Text)

  • Publication Type: Conference Paper / Full Text
  • Volume: 1
  • City: İstanbul
  • Country: Turkey
  • Page Numbers: pp.470-485
  • Karadeniz Technical University Affiliated: Yes

Abstract

Possessing knowledge about the inclination (αcr) of potential failure surface, the magnitude (Pae) and application point (zae) of active earth thrust guides engineers for the proper dimensioning of retaining walls under static and seismic conditions. The study aims to use Gaussian Process Regression (GPR) as an alternative method, considering 4 covariance functions and 2 data sampling techniques, that predicts the three design parameters taking into account wall height, surcharge load, cohesion and internal friction angle of backfill soil, vertical and horizontal seismic acceleration coefficients, adhesion, and tension cracks problem for a retaining wall in the case of broken terrain. For this purpose, the two datasets are employed to build the predictive mathematical models and then to verify the usability of the GPR models, respectively. The accuracy and efficiency of the GPR models are evaluated by comparison with the results of an enlightening article using the six statistical indicators (e.g. R2, R2adj, MSE, RMSE, MAE, and MAPE). The statistical indicators show that GPR is a very effective machine learning algorithm to estimate Pae and αcr. However, GPR indicates a slightly weaker performance for zae. This study also involves a statistical benchmarking to achieve the superior advantage of GPR with Support Vector Regression (SVR).