Performance evaluation of machine learning methods for ground settlement prediction

Šerifović-Trbalić, Amira (author), Prljača, Naser (author), Paparo, Ausilia (author), Lorusso, Martin (author)

Prediction of tunneling-induced ground settlements is an important task during tunnel excavation in urban areas. Ground settlements should be limited within a tolerable threshold to avoid damages to existing buildings and infrastructures during and after the construction. Machine learning (ML) methods have been gaining an increasing popularity in many fields, including tunnel excavations, as a powerful learning and predicting technique. The paper analyzes the possibilities of different machine learning methods to predict the ground surface settlement induced by tunneling. Three different ML approaches, including support vector regression (SVR), multilayer perceptron (MLP), and long short-term memory (LSTM) networks, are utilized. Two techniques are used for the hyperparameter optimization: particle swarm optimization (PSO) and grid search (GS) methods. To assess the performance of the ML methods, three performance metrics are used: the mean absolute error (MAE), root mean square error (RMSE) and mean absolute percentage error (MAPE). The paper demonstrates the applicability of the three ML methods in tunneling-induced ground settlement prediction for real-world settlement datasets. The obtained experimental results indicate that the proposed ML models can accurately and efficiently predict the ground settlement.

• Language: English
• Work type: Article
• Typology: 1.01 - Original Scientific Article
• Organization: FE - Faculty of Electrical Engineering
• Year: 2025
• Number of pages: Str. 13-25
• Numbering: Letn. 92, št. 1/2
• PID: 20.500.12556/RUL-183227
• UDC: 004.85:69
• ISSN on article: 0013-5852
• COBISS.SI-ID: 280328451
• Publication date in RUL: 08.06.2026

Record is a part of a journal

• Title: Elektrotehniški vestnik
• Publisher: Strokovna zadruga koncesijoniranih elektrotehnikov, Elektrotehniška zveza Slovenije
• ISSN: 0013-5852
• COBISS.SI-ID: 742916

Licences

• License: CC BY 4.0, Creative Commons Attribution 4.0 International
• Link: http://creativecommons.org/licenses/by/4.0/
• Description: This is the standard Creative Commons license that gives others maximum freedom to do what they want with the work as long as they credit the author.

Secondary language

• Language: Slovenian
• Title: Ocena činkovitosti metod strojnega učenja za napoved posedanja tal

Abstract

Napoved posedanja tal, ki ga povzroča gradnja predorov, je ključnega pomena pri izkopavanju predorov v urbanih območjih. Posedanje tal mora ostati znotraj sprejemljivih mejnih vrednosti, da se preprečijo poškodbe obstoječih stavb in infrastrukture med gradnjo in po njej. Metode strojnega učenja pridobivajo vse večjo priljubljenost na različnih področjih, vključno z gradnjo predorov, saj omogočajo učinkovito učenje in napovedovanje. Prispevek analizira možnosti uporabe različnih metod strojnega učenja za napoved posedanja tal, ki ga povzroča gradnja predorov. Uporabljeni so trije pristopi strojnega učenja: regresija s podporo vektorjev, večplastni perceptron in nevronske mreže dolgega kratkoročnega spomina. Za optimizacijo hiperparametrov sta uporabljeni dve tehniki: optimizacija z rojem delcev in metoda iskanja po mreži. Za oceno učinkovitosti metod strojnega učenja so uporabljene tri metrike: povprečna absolutna napaka, kvadratna srednja napaka in povprečna absolutna odstotkovna napaka. Prispevek prikazuje uporabnost treh metod strojnega učenja za napoved posedanja tal na realnih podatkovnih zbirkah. Eksperimentalni rezultati kažejo, da predlagani modeli strojnega učenja omogočajo natančno in učinkovito napoved posedanja tal

• Keywords: posedanje tal, gradnja predorov, urbana območja, strojno učenje