Settlement prediction of micropile supported raft using machine learning: modelling and performance evaluation

Micropiles are composed of steel bars and grouts which can effectively address the challenges of constructing medium-height and medium-weight structures on deep, soft clayey soil deposits. Micropile supported raft foundations not only regulate total as well as differential settlement but also boost the capacity. The accurate prediction of settlement for a micropile supported raft foundation is crucial due to diverse soil properties, nature and placement of loading and complex soil foundation interaction. For this purpose, a machine learning (ML) based settlement prediction model is proposed in this paper. A dataset consisting of 350 data points with 13 distinct features were collected from static vertical compressive load test conducted on micropile supported raft foundation. Micropile was constructed as a cast in-situ gravity grouted with neat cement. To identify the statistically significant features, various machine learning models were used to accurately predict the settlement of micropile-supported rafts. The evaluation of the proposed regression models showed that the Extreme Gradient Boosting algorithm achieved the best results, with metrics including MAE = 1.25, R² = 0.98, RMSE = 2.36, MAPE = 0.56, VAF = 98.18, RSR = 0.13, and an average PI width of 8.96. These results highlight the model’s high efficiency. Further the workability of the proposed model was also justified by implementing prevailing literature data which substantiate model’s reliability. Therefore, the proposed methodology can be implemented for real-time settlement prediction of micropile-supported rafts.