Hybrid Random Forest-Based Models for Earth Pressure Balance Tunneling-Induced Ground Settlement Prediction

Construction-induced ground settlement is a serious hazard in underground tunnel construction. Accurate prediction has great significance ensuring the surface building’s stability and human safety. To that end, 148 sets of data were collected from Singapore Circle Line rail traffic project containing seven defining parameters to create database for predicting settlement. These are depth (H), advance rate (AR), EPB earth pressure (EP), mean SPTN value soil crown (Sm), water content layer (MC), modulus elasticity (E), grout used injecting into tail void (GP). Three hybrid models consisting random forest (RF) three types meta-heuristics, Ant Lion Optimizier (ALO), Multi-Verse Optimizer (MVO), Grasshopper Optimization Algorithm (GOA), developed predict Furthermore, absolute error (MAE), percentage (MAPE), coefficient determination (R2) root square (RMSE) assess predictive performance constructed The evaluation results demonstrated GOA-RF with population size 10 achieved most outstanding capability indices MAE (Training set: 2.8224; Test 2.3507), MAPE 40.5629; 38.5637), R2 0.9487; 0.9282), RMSE 4.93; 3.1576). Finally, sensitivity analysis indicated MC, AR, Sm, GP have significant impact on based model.