Prediction of Axial Bearing Capacity of Piles using Sophisticated ML Algorithms Tuned through Random and Grid Search

Abstract

Pile foundations support structures by transferring loads to deep sub-surface strata, designed to bear the maximum design load without failure. Recent studies are focused on developing innovative models to estimate the pile capacity on efficient ground in less time. The pile load tests are difficult to perform and time-consuming. So, this study aims to address existing gaps in geotechnical engineering research, specifically in pile strength estimation, by deploying advanced machine learning algorithms, namely Random Forest (RF), Support Vector Regression (SVR), and Xtreme Gradient Boost (XG Boost), which are meticulously fine-tuned using hyperparameter optimization techniques, such as Grid Search (GS) and Random Search (RS). The models were formulated in a high-level programming language, namely Python. The model's efficacy was assessed through Root Mean Square Error (RMSE), Coefficient of Determination (R2), and Standard Deviation (SD). The test results show that each model performs well; however, the XGBoost algorithm shows higher efficacy, with high accuracy on the data sets (R2 = 0.933).