Ground settlement prediction by NATM with ensemble learning

The primary objective of this study is to develop a machine learning model capable of predicting tunnel-induced settlement resulting from excavation processes. To achieve this, a set of finite element – based datasets was generated from rigorously calibrated and validated numerical models. These datasets were then employed to construct an ensemble learning model for settlement prediction. The input variables were selected to include key geotechnical and geometric factors, specifically the surrounding soil properties and tunnel configuration parameters, which collectively govern the deformation behaviour during excavation. The selected soil parameters comprise cohesion, internal friction angle, Poisson’s ratio, and the elastic modulus. The tunnel-related parameters include tunnel diameter, depth of cover, and excavation method. The maximum surface settlement above the tunnel was designated as the output variable. Two machine learning models – Gradient Boosting and XGBoost – were developed for this purpose. The advantage of employing these two algorithms lies in their complementary strengths, where their integration helps to mitigate the individual limitations of each model especially in the case of relatively small data-set. The findings confirm the effectiveness of the proposed approach in predicting excavation-induced tunnel settlement, demonstrating its practical utility for risk assessment and mitigation planning.