Artificial neural networks for predicting the maximum surface settlement induced by EPB-TBM: The Algiers Metro Case

This study develops an Artificial Neural Network (ANN) model to predict the maximum surface settlement induced by Earth Pressure Balance Tunnel Boring Machines (EPB-TBM), with application to the Algiers Metro Line 1 extension. The model was trained on a dataset of 442 samples integrating operational parameters, geometric features, and geotechnical measurements. Following a structured preprocessing phase, the ANN achieved high predictive accuracy, with coefficients of determination exceeding 0.92 and a root mean square error below 4.2 mm. In comparison to Random Forest and XGBoost, the ANN demonstrated superior performance in capturing nonlinear relationships. Sensitivity analysis identified equivalent cover depth, groundwater head, and distance from the station as the most influential variables. These results highlight the model’s reliability as a data-driven tool for settlement prediction and tunneling risk management in urban environments.