Intelligent support system for ground settlement management during TBM tunneling by combining machine learning with statistical analysis

Intelligent support system for ground settlement management during TBM tunneling by combining machine learning with statistical analysis | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Intelligent support system for ground settlement management during TBM tunneling by combining machine learning with statistical analysis Kibeom Kwon, Minkyu Kang, Dongku Kim, Khanh Pham, Hangseok Choi This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4771476/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Ground settlement management is crucial in tunnel boring machine (TBM) tunneling. Previous studies on predicting ground settlement have required substantial assumptions or information, making it challenging to explicitly determine their predictive criteria. This study developed an intelligent TBM operation support system for ground settlement management, by combining machine learning and statistical analysis. Initially, measured settlements were categorized into three classes: heaving, normal, and large settlement. Based on three key features selected through correlation analysis, the predictive criterion for the heaving class was determined using an initial model based on a decision tree algorithm. Subsequently, through statistical analysis, the predictive criteria for the normal and large settlement classes were established. The developed support system, using only three key features, achieved an accuracy of 0.847, F1 score of 0.784, precision of 0.798, and recall of 0.770, outperforming two ensemble machine learning models that used nine features. Moreover, the system can provide explicit predictive criteria, enhancing its practical applicability. Error analysis revealed that among the four instances misclassified by the support system, three pertained to minor settlements within ± 3 mm. Physical sciences/Engineering/Civil engineering Physical sciences/Mathematics and computing/Scientific data Intelligent support system Tunnel boring machine Ground settlement Machine learning Statistical analysis Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 1. Introduction The increasing population density and rising social costs associated with traffic congestion have driven a growing demand for tunnel construction [ 1 ]. Among various tunnel construction methods, tunnel boring machines (TBMs) have been extensively adopted owing to their superior eco-friendliness, stability, and constructability. However, the inherent geological, mechanical, and combined uncertainties encountered during TBM tunneling can result in adverse accidents, such as collapse and water/mud inflow [ 2 ]. Among these accidents, managing ground settlement within acceptable limits is crucial to prevent severe damage to existing surface and underground structures [ 3 , 4 ]. Therefore, it is essential to adjust TBM operation conditions based on a reliable support system which effectively predicts ground settlements. Previous studies on predicting ground settlement induced by tunneling have employed three primary approaches: empirical-statistical [ 5 , 6 ], analytical [ 7 – 9 ], and numerical [ 10 – 12 ] methods. The empirical-statistical approach employs simple mathematical formulas based on prior tunnel excavation experience and data collected from extensive field observations. This method has been widely used in practical applications [ 13 ]. However, applying this approach to TBM tunneling is challenging because it doesn't easily account for TBM operation features like thrust force and torque. The analytical approach provides closed-form solutions for predicting induced ground settlement through theoretical deductions. Despite its theoretical foundation, deriving suitable solutions using this approach is inherently complex due to necessary assumptions (e.g., isotropic, homogeneous, and incompressible soil) and uncertain relationships between settlement-inducing features, such as soil-structure interactions [ 14 , 15 ]. Furthermore, most analytical results cannot capture TBM operation features, similar to empirical results [ 16 ]. With the rapid advancement of computer technology and numerical algorithms, several studies have utilized numerical approaches to predict ground settlement [ 17 , 18 ]. Unlike previous methods, numerical approaches can account for TBM operation and site-specific features (e.g., geology and geometry). However, they have limitations stemming from the use of constitutive models with substantial assumptions about uncertain material properties, operation features, and boundary conditions [ 19 ]. Moreover, considerable computational resources are needed to ensure the reliability of obtained results. Recently, machine learning approaches have been introduced in various studies to predict TBM tunneling-induced ground settlement [ 18 , 19 , 20 – 26 ]. This approach can consider TBM operation and site-specific features using extensive data and effectively analyze complex relationships between settlement-inducing features. However, most of machine learning-based studies have relied on black box-type algorithms, making it challenging to provide explicit predictive criteria. Furthermore, these algorithms may have limited practical applicability due to the need for extensive or localized feature information. This study aims to develop an intelligent TBM operation support system for ground settlement management. Correlation analysis is conducted to identify three key features utilized in the support system. The predictive criteria for three settlement classes are established through a combination of machine learning and statistical analysis. To validate the practical applicability of the developed system, a comparative analysis is conducted against ensemble machine learning models. Furthermore, error analysis is performed to evaluate the prediction performance of the support system regarding ground settlement values. 2. Overview of machine learning algorithms 2.1. Decision tree (DT) Decision tree (DT) divides the predictor space into multiple regions, resulting in a tree-like structure developed based on a set of splitting rules [ 27 ]. This structure comprises various nodes, including root, internal, and leaf nodes. A root node (containing the entire dataset) is split into two internal nodes (each containing a split dataset) based on an optimal feature that maximizes the information gain for a given index (e.g., the Gini index). Subsequently, the internal nodes are recursively divided until a leaf node is reached. In classification problems like this study, the final prediction is made using a set of these splitting rules. Unlike black box-type algorithms, DT is transparent as it explicitly provides the predictive criteria [ 28 ]. DT was used to develop an initial model in this study. 2.2. Random forest (RF) Random forest (RF) is a bagging-type ensemble machine learning algorithm introduced by [ 29 ] based on a set of decision trees. For each decision tree construction, a bootstrap sample (i.e., a randomly selected sample from the original training set with replacement) is drawn, with the number of data instances and features in the sample being randomly determined. The decision tree then evolves based on the selected bootstrap sample. In this manner, n bootstrap samples and their corresponding n decision trees are repeatedly constructed. By aggregating all the decision tree outputs, the final prediction is determined as the mode (in classification) or the average value (in regression). Because of its high prediction performance and low risk of overfitting, RF has been widely applied in various fields. 2.3. Extreme gradient boosting (XGB) Extreme gradient boosting (XGB) is a gradient boosting-type ensemble learning algorithm extensively applied in diverse domains, initially proposed by [ 30 ]. The core of gradient boosting-type algorithms lies in the sequential creation of numerous weak learners, with each learner focused on rectifying the residual (i.e., the gradient of a loss function) left by its predecessor. The final strong learner emerges as an ensemble of these weak learners. Specifically, the predictions of these weak learners are then weighted based on their individual performance, and the aggregated prediction of these weighted predictions forms the outcome of the strong learner. Nevertheless, gradient boosting-type algorithms run the risk of constructing an excessively complex model by exclusively minimizing the residual, potentially leading to overfitting. To mitigate this risk, XGB incorporates an advanced objective function that combines the loss function with a regularization term, optimally balancing prediction performance and model complexity. In this study, RF and XGB were used to develop comparison models for validating the support system. 3. Dataset creation 3.1. Project overview This study used a dataset obtained from a slurry shield TBM tunneling project in Hong Kong. Twin-bored slurry shield TBM subway tunnels with a total length of 850 m, comprising both up-track and down-track tunnels, were constructed at the tunneling site. The up-track tunnel was excavated five months after the down-track tunnel to minimize ground disturbance. These tunnels were categorized as shallow tunnels because their cover depth was less than twice the TBM excavation diameter [ 31 ]. The detailed specifications of the slurry shield TBM are summarized in Table 1 . Table 1 Slurry shield TBM specifications. Description Specification TBM excavation diameter 7.4 m TBM length 11.45 m Max. thrust force 47,897 kN Max. torque 5 MN⋅m Opening ratio 29% Segment diameter 7.1 m (OD), 6.5 m (ID) Segment width 1.5 m After the TBM passed through the second tunnel (i.e., up-track tunnel), daily settlement measurements were made at 253 locations along the tunnel alignment. The final settlement value at each location was determined as the average over five days, typically around the 20th day following the TBM passage. Four different ground types were encountered in the up-track alignment: fill, alluvium, completely decomposed granite (CDG), and highly decomposed granite (HDG). HDG and CDG can be classified as highly fractured normal to weak rock and very weak rock to weathered soil, respectively [ 31 ]. Representative geotechnical properties of these ground types are summarized in Table 2 . The longitudinal geological profile along the tunnel alignment is illustrated in Fig. 1 . More detailed information on data acquisition procedures and ground types can be referred to [ 24 , 25 ]. Table 2 Geotechnical properties of the encountered ground types [ 25 ]. Ground type N-value Cohesion [kPa] Friction angle [ \(\:^\circ\:\) ] Permeability [m/s] Fill 10 0 35 3.15 \(\:\times\:\) 10 -5 Alluvium D * \(\:<\) 4: 10 D \(\:\ge\:\) 4: 1.67 D + 3.33 0 35 2.32 \(\:\times\:\) 10 -5 CDG D \(\:<\) 5: 15 D \(\:\ge\:\) 5: 2.71 D + 1.43 8 38 3.34 \(\:\times\:\) 10 -6 HDG 200 12 40 4.60 \(\:\times\:\) 10 -6 * D: Depth [m] 3.2. Data pre-processing Generally, features that induce settlement are categorized into three categories: geometry, geology, and TBM operation characteristics [ 20 ]. In this study, the features commonly addressed and readily obtainable in various TBM tunneling projects were selected to ensure the practical applicability of the developed support system [ 23 , 26 ]. Previous studies considered the cover depth and the horizontal distance between the tunnel centerline and settlement monitoring points (HTM) as geometry features [ 23 – 26 , 32 , 33 ]. These geometry features become particularly important for predicting settlement in shallow tunnels, similar to the site of this study. Geology features adopted here include the standard penetration test (SPT) N-value and the distance between the ground surface and the groundwater level (DSG). These features account for the potential influence of ground strength and groundwater on settlement. Notably, the SPT N-value, being more subdivided than other geotechnical properties (refer to Table 2 ), is likely to influence machine learning predictions [ 26 ]. Concurrently, this study focused on five commonly measured TBM operation features in tunneling projects: thrust force, torque, face pressure, penetration rate, and grouting volume. These features have demonstrated their contributions to ground settlement in several previous studies [ 25 , 34 – 38 ]. Since settlement measurements were conducted after the excavation of the up-track tunnel, this study used the features in each category, corresponding to the up-track tunnel excavation. Although, in certain chainages, settlements were measured at various locations with different horizontal distances. In this study, the settlements closest to the tunnel centerline were exclusively considered. Statistical descriptions of the settlement-inducing features and the settlements are presented in Table 3 . Table 3 Statistical description of the selected features and settlements. Category Feature Min Q1 Median Q3 Max COV * Unit Geometry Cover depth 6.70 7.10 7.50 8.90 12.80 0.17 m HTM 0.002 4.206 10.118 16.413 39.892 0.82 m Geology N-value 10.00 15.00 16.00 22.00 36.00 0.36 - DSG 3.02 3.88 4.31 4.68 7.23 0.23 m Operation Thrust force 8,676 12,850 13,700 15,400 26,000 0.24 kN Torque 0.20 0.50 0.70 1.00 1.80 0.42 MN∙m Face pressure 1.20 1.45 1.59 1.84 2.45 0.16 bar Penetration rate 13.00 27.00 35.00 40.00 48.00 0.25 mm/min Grouting volume 5.30 6.30 6.43 6.60 7.60 0.05 m 3 - Settlement -6.90 2.50 4.40 8.00 16.00 0.63 mm * COV is the coefficient of variation, denoted as the standard deviation normalized by the mean. This study categorized the measured settlements ( \(\:S\) ) into three classes based on their magnitude: heaving ( \(\:S<0\:mm\) ), normal ( \(\:0\:mm10\:mm\) ), as summarized in Table 4 . A total of 129 instances comprises nine features (including geometry, geology, and operation features) and one target variable (settlement class). Table 4 Categorized settlement classes defined in this study. No Class Count Proportion 1 Heaving ( \(\:\varvec{S}<0\:\varvec{m}\varvec{m}\) ) 16 12.4% 2 Normal ( \(\:0\:\varvec{m}\varvec{m}10\:\varvec{m}\varvec{m}\) ) 21 16.3% 3.3. Data analysis Identifying highly correlated features with the categorized settlement class among all the collected features is important. To simplify the developed support system for practical use by TBM operators, only a small number of features were selectively adopted. Through correlation analysis, the relationships between these features and the target variable (i.e., the settlement class) were identified. The results of this correlation analysis, presented by Pearson’s coefficient, are shown in Fig. 2 . The correlation analysis indicated that DSG, thrust force, torque, and face pressure were the most relevant features with respect to the settlement class. Specially, thrust force and torque exhibited strong negative correlations with the settlement class, represented by Pearson’s coefficients of -0.49 and − 0.48, respectively. In simpler terms, increases and decreases in these features (i.e., thrust force and torque) were closely associated with the heaving and large settlement classes, respectively. This observation aligns with findings from previous studies [ 18 , 22 , 39 ]. Consequently, thrust force and torque were selected as the key features for the support system. Face pressure has been a critical parameter for controlling induced ground settlements due to its significant influence on face stability during TBM tunneling [ 18 ]. Figure 2 reinforces this perspective by demonstrating a strong correlation between face pressure and the settlement class, indicated by a Pearson’s coefficient of -0.44. Nevertheless, it is important to note that face pressure also exhibited a strong correlation with thrust force (i.e., Pearson’s coefficient ≥ 0.7) and had a low coefficient of variation (COV), as shown in Table 3 . Similarly, despite DSG showing a high Pearson’s coefficient of -0.50, it was also strongly correlated with thrust force. Moreover, recording DSG can be relatively uncertain since it was interpolated from results derived from various exploration boreholes. Consequently, to enhance the practical applicability of the support system, this study replaced these two features, face pressure and DSG, with thrust force. In the support system development, most of the features from Table 3 , which demonstrated weak correlations with the settlement class, were excluded. However, even though the correlation between N-value and the settlement class was low, this study retained N-value as a geology feature. This decision was made to account for the significance of weak ground, which typically corresponds to a high risk of ground settlement [ 40 ]. As a result, the dataset was streamlined to consist of three features (i.e., thrust force, torque, and N-value) along with the target variable (i.e., the settlement class). 4. System development 4.1. Criterion for heaving class The primary objective of this study is to develop an intelligent TBM operation support system with ground settlement prediction induced by TBM tunneling. This system classifies settlement into three classes, as outlined in Table 4 , using three key features: thrust force, torque, and N-value. However, establishing predictive criteria for this system from the beginning, especially when dealing with four-dimensional data encompassing thrust force, torque, N-value, and settlement class, is a complex challenge. Therefore, this study first constructed an initial model employing a machine learning approach to provide preliminary predictive criteria. The initial model was produced using the DT algorithm, chosen for its capacity to offer clear visualization and an intuitive interpretation of predictive criteria, in contrast to black box-type algorithms such as the RF and XGB algorithms. Subsequently, the dataset was divided into training and test sets, with 80% of the instances allocated for training and the remaining 20% for testing. It is important to note that the same test set was used for evaluating prediction accuracies corresponding to the initial model, the support system, and comparison models. In addition, this study utilized Bayesian optimization (BO) combined with 5-fold cross-validation for hyperparameter tuning. BO is an efficient strategy for finding optimal solutions, such as hyperparameter combinations that maximize an unknown objective function (e.g., accuracy and coefficient of determination). It relies on two core components: surrogate model and acquisition function. The surrogate model estimates the posterior distribution of the objective function using prior knowledge derived from previously evaluated samples (e.g., hyperparameter combinations) based on Bayes’ theorem [ 41 ]. The acquisition function then leverages this estimated distribution to determine the next sample for evaluation [ 42 ]. In this study, a Gaussian process (GP) was employed as the surrogate model, while expected improvement (EI) were adopted as the acquisition function. Detailed information on BO based on GP and EI can be found in [ 43 ]. Table 5 presents the explored hyperparameters and their corresponding search spaces. It can be noted that the max_depth hyperparameter was set to less than five to mitigate overfitting and reduce complexity. Table 5 Search spaces for the explored DT hyperparameters. Hyperparameter Description Search space max_depth Maximum depth of tree model 2–4 min_samples_split Minimum number of samples required to split a node 3–10 min_samples_leaf Minimum number of samples required in a leaf 3–10 The initial model was developed with the following optimal hyperparameters: max_depth = 4, min_samples_split = 10, and min_samples_leaf = 3. Table 6 and Fig. 3 show the prediction performance in the training and test phases. In this study, four performance metrics were used: accuracy, F1 score, precision, and recall. Macro-averaging precision and recall were adopted, where macro-averaging assigns equal weight to each class, regardless of the number of instances in each class [ 44 ]. Table 6 Prediction performance for each class obtained by the initial model. Performance metric Training Test Accuracy 0.825 0.808 F1 score 0.751 0.732 Precision 0.858 0.783 Recall 0.668 0.687 Although over two-thirds of the heaving class in both training and test sets were correctly predicted, significant prediction errors were observed in the test phase for the large settlement class, with the recall for that class of 0.500. Throughout the training and test phases, only eight out of the 21 cases for the large settlement class were accurately predicted, while the rest were misclassified as the normal class. This misclassification issue can be attributed to the initial model’s lack of emphasis on predicting the large settlement class, which is a common challenge when applying machine learning algorithms to imbalanced datasets. Consequently, according to the DT structure of the initial model (Fig. 4 ), the predictive criterion for the heaving class (Eq. ( 1 )) was adopted for the support system developed in this study. However, additional analyses were required to address the misclassifications between the normal and large settlement classes, thereby deriving their predictive criteria. $$\:N\_value\le\:28\:and\:Thrust\:force>\text{17,695}\:kN$$ 1 4.2. Criterion for large settlement class In this study, statistical analysis was employed to establish the criteria for predicting the normal and large settlement classes. Typically, in the statistical analysis of dataset distributions, five key indices are considered: the minimum and maximum values, the lower and upper quartiles (Q1 and Q3), and the median. These indices of three features (thrust force, torque, and N-value) are summarized for each settlement class in Table 7 and visually depicted in a box plot shown in Fig. 5. Table 7 Statistical description of the selected features corresponding to each class. Indices Thrust force [kN] Torque [MN∙m] N-value Heaving Normal Large settlement Heaving Normal Large settlement Heaving Normal Large settlement Min 14,500 9,628 8,822 0.5 0.2 0.3 16 10 10 Q1 17,890 12,700 12,000 1.0 0.5 0.325 17 15 15.25 Median 19,500 13,500 13,100 1.2 0.7 0.6 22 15 17.5 Q3 20,600 14,800 14,444 1.3 0.9 0.775 22 19 24.75 Max 25,800 26,000 15,500 1.5 1.5 1.0 27 34 31 According to Fig. 5, it becomes evident that both thrust force and torque decrease as settlement magnitude increases, following the order of the heaving, normal, and large settlement classes. While the decline in thrust force is less pronounced, the trend in torque is more noticeable. Hence, this study defined the criterion for predicting the large settlement class as follows: instances that do not meet the heaving class criterion while concurrently exhibiting torque values below the lower quartile (Q1) for the normal class. Based on Eq. ( 1 ) and Table 7 , the criterion for predicting the large settlement class is summarized in Eq. ( 2 ). $$\:[N\_value>28\:or\:Thrust\:force\le\:\text{17,695}\:kN]\:and\:[Torque<0.5\:MN\cdot\:m]$$ 2 4.3. Criterion for normal class The criterion for predicting the normal class can be determined as follows: instances that do not meet the conditions specified in equations ( 1 ) and ( 2 ). Therefore, a data-driven support system for predicting ground settlement, comprising three criteria for predicting each settlement class, was formulated by combining machine learning with statistical analysis. Figure 6 illustrates the predictive criteria within the developed support system. 5. System application 5.1. Prediction performance The prediction performances of the developed support system were assessed using the test set, comprising a total of 26 instances, as described in Section 3.2 . Table 8 presents the prediction performances of both the initial model and the support system when evaluated on the same test set, while Fig. 7 illustrates a confusion matrix for the developed support system. Table 8 Prediction performance of the initial model and the developed support system. Performance metric Initial model Support system Accuracy 0.808 0.847 F1 score 0.732 0.784 Precision 0.783 0.798 Recall 0.687 0.770 According to Table 8 , the support system enhanced prediction performance compared to the initial model, achieving an accuracy of 0.847, an F1 score of 0.784, a precision of 0.798, and a recall of 0.770. Notably, in the large settlement class, the support system correctly predicted three out of four instances, whereas the initial model correctly predicted only two. Meanwhile, the number of correctly predicted instances for the heaving and normal classes remained consistent. Importantly, all incorrectly predicted instances were classified into adjacent classes, reducing the risk of applying conflicting countermeasures. Consequently, the support system, which integrates machine learning with statistical analysis, outperformed the DT-based initial model, while both systems can explicitly provide predictive criteria. 5.2. Comparative analysis The purpose of this section is to validate the practical applicability of the developed support system through a comparative analysis with different comparison models. Typically, bagging and boosting-based ensemble machine learning algorithms have been widely employed to address imbalanced datasets [ 45 , 46 ]. As shown in Table 4 , a substantial proportion of the dataset corresponds to the normal class, signifying its imbalance. Consequently, this study developed two comparison models using representative bagging and boosting-based ensemble machine learning algorithms, namely RF and XGB. It can be noted that all the features listed in Table 3 were used in building the comparison models to highlight the simplicity of the developed support system that employs only three features. For hyperparameter tuning of these comparison models, BO and 5-fold cross-validation were employed. The search spaces for RF and XGB hyperparameters are described in Table 9 . Table 9 Search spaces of RF and XGB hyperparameters. Algorithm Description Search space RF n_estimators 100–1000 max_depth 2–30 min_samples_split 3–10 min_samples_leaf 3–10 XGB n_estimators 100–1000 max_depth 2–30 min_child_weight 0–5 gamma 0–1 subsample 0–1 learning-rate 0.01–1 Table 10 presents the prediction performances of the developed support system and the comparison models (i.e., RF and XGB models) when evaluated on the same test set, while Fig. 8 illustrates the confusion matrices for each model. Table 10 Prediction performance of the support system and comparison models. Performance metric Support system RF model XGB model Accuracy 0.847 0.808 0.769 F1 score 0.784 0.715 0.649 Precision 0.798 0.680 0.630 Recall 0.770 0.753 0.670 The comparative analysis demonstrated that all the performance metrics of the support system surpassed those of the RF and XGB models. This suggests that the support system, which combines machine learning with statistical analysis, is more effective in predicting ground settlement than the ensemble machine learning algorithms. Moreover, the practical applicability of the support system was validated, as it was developed using only three features (i.e., thrust force, torque, and N-value) and provided explicit predictive criteria, as proposed in this study. The overall flow chart outlining the intelligent support system developed in this study is presented in Fig. 9 . 5.3. Error analysis for the support system This study performed error analysis examining the prediction results for the instances within the test set, with the consideration of their settlement values. Figure 10 displays the settlement values of these instances and indicates whether the developed support system predicted them correctly or incorrectly. As shown in Fig. 10 , it was evident that three out of four incorrectly predicted instances had settlement values within \(\:\pm\:\) 3 mm. These inaccuracies, closely associated with subtle ground deformations, were expected to have a relatively minor impact on the safety and efficiency of TBM tunneling projects. In contrast, it should be noted that the instance characterized by a substantial settlement value of 16 mm was incorrectly predicted, which could have significant adverse consequences. As shown in Table 3 , this instance represented the maximum settlement value in the entire dataset. It can be inferred that the developed support system faced challenges in accurately predicting this instance due to the scarcity of large settlement data, particularly those exceeding 15 mm. This observation highlights the need for an extensive dataset encompassing a substantial number of instances with large settlement values. Such a dataset would not only enhance the prediction accuracy of the developed support system but also ensure the safety and efficiency of TBM tunneling projects. 5.4. Limitation of this study Despite achieving the support system with superior prediction performances and practical applicability, this study has a limitation. The support system was constructed using a relatively small dataset of only 129 instances obtained from a single specific tunneling site, potentially limiting its generalization capability. Therefore, future studies could focus on building an integrated database from multiple TBM tunneling projects with similar working conditions to enhance and complement the support system’s performance. 6. Conclusions This study introduced an intelligent TBM operation support system for ground settlement management, which combined machine learning with statistical analysis. The practical applicability of the support system was validated by comparison with ensemble machine learning models. The principal findings of this study are summarized as follows. The support system incorporated three predictive criteria employing three key features: thrust force, torque, and N-value. First, a DT-based initial model was employed to establish the predictive criterion for heaving. Subsequently, statistical analysis was used to determine the predictive criterion for large settlement. Finally, the predictive criterion for the normal class involved instances that did not meet the preceding criteria. The support system, employing just three key features, outperformed RF and XGB models that used nine features. Moreover, the support system provides explicit predictive criteria, unlike the comparison models. These advantages underscore the practical applicability of the developed support system. Out of the four instances incorrectly predicted by the support system, three were associated with settlement values within ± 3 mm, which generally have a minor impact on TBM tunneling. Conversely, the remained one instance had a substantial settlement value of 16 mm, indicating the need for support system improvement by incorporating a more extensive dataset, particularly encompassing significantly large settlement data. Declarations Data availability The datasets generated and/or analyzed during the current study are not publicly available due to site information security issues but are available from the corresponding author on reasonable request. Competing Interests The authors declare no competing interests. Acknowledgements This research was supported by the National R&D Project for Smart Construction Technology (RS-2020-KA157074) and for Consecutive Excavation Technological Development Project of Tunnel Boring Machine (RS-2022-00144188) funded by the Korea Agency for Infrastructure Technology Advancement under the Ministry of Land, Infrastructure, and Transport. Contributions Kibeom Kwon: Conceptualization, Methodology, Writing - Original Draft. Minkyu Kang: Data Curation, Formal analysis, Validation. Dongku Kim: Data Curation, Software, Visualization. Khanh Pham: Software, Formal analysis, Methodology. Hangseok Choi: Conceptualization, Supervision, Writing - Review & Editing. References Kwon, K, Kang, M, Kim, D, & Choi, H. Prioritization of hazardous zones using an advanced risk management model combining the analytic hierarchy process and fuzzy set theory. Sustainability 15(15), 12018 (2023) Sousa RL, Einstein HH. Lessons from accidents during tunnel construction. Tunn. Undergr. Space Technol. 113, 103916 (2021) Ercelebi, S. G, Çopur, H., & Ocak, I. Surface settlement predictions for Istanbul Metro tunnels excavated by EPB-TBM. Environ. Earth. Sci. 62, 357–365 (2011) Chung, H. et al. A causal network-based risk matrix model applicable to shield TBM tunneling projects. Sustainability 13(9), 4846 (2021) Peck, R. B. Deep excavations and tunnelling in soft ground. Proceedings of the 7th International Conference on Soil Mechanics and Foundation Engineering, Mexico City, Mexico, 225–290 (1969) Atkinson, J. H. & Potts, D. M. Subsidence above shallow tunnels in soft ground. J. Geotech. Engrg. Div. 103(4), 307–325 (1977) Sagaseta, C. Analysis of undrained soil deformation due to ground loss. Geotechnique 37(3), 301–320 (1987) Loganathan, N., & Poulos, H. G. Analytical prediction for tunneling-induced ground movements in clays. J. Geotech. Geoenviron. Eng. 124(9), 846–856 (1998) Verruijt, A. A complex variable solution for a deforming circular tunnel in an elastic half-plane. Int. J. Numer. Anal. Methods Geomech. 21(2), 77–89 (1997) Franzius, J. N., Potts, D. M., & Burland, J. B. The influence of soil anisotropy and K0 on ground surface movements resulting from tunnel excavation. Geotechnique 55(3), 189–199 (2005) Chakeri, H., Ozcelik, Y., & Unver, B. Effects of important factors on surface settlement prediction for metro tunnel excavated by EPB. Tunn. Undergr. Space Technol. 36, 14–23 (2013) Huang, H. et al. Simplified procedure for finite element analysis of the longitudinal performance of shield tunnels considering spatial soil variability in longitudinal direction. Comput. Geotech. 64, 132–145 (2015) Moghaddasi, M. R., & Noorian-Bidgoli, M. ICA-ANN, ANN and multiple regression models for prediction of surface settlement caused by tunneling. Tunn. Undergr. Space Technol. 79, 197–209 (2018) Zhou, J. et al. Feasibility of random-forest approach for prediction of ground settlements induced by the construction of a shield-driven tunnel. Int. J. Geomech. 17(6): 04016129 (2017) Kim, D., Pham, K., Park, S., Oh, J. Y., & Choi, H. Determination of effective parameters on surface settlement during shield TBM. Geomech. Eng. 21(2), 153–164 (2020) Chen, R. P., Song, X., Meng, F. Y., Wu, H. N., & Lin, X. T. Analytical approach to predict tunneling-induced subsurface settlement in sand considering soil arching effect. Comput. Geotech. 141, 104492 (2022) Kim, K., Oh, J. Y., Lee, H., Kim, D., & Choi, H. Critical face pressure and backfill pressure in shield TBM tunneling on soft ground. Geomech. Eng. 15(3), 823–831 (2018) Zhou, X., Zhao, C., & Bian, X. Prediction of maximum ground surface settlement induced by shield tunneling using XGBoost algorithm with golden-sine seagull optimization. Comput. Geotech. 154, 105156 (2023) Ye, X. W., Jin, T., & Chen, Y. M. Machine learning-based forecasting of soil settlement induced by shield tunneling construction. Tunn. Undergr. Space Technol. 124, 104452 (2022) Suwansawat, S., & Einstein, H. H. Artificial neural networks for predicting the maximum surface settlement caused by EPB shield tunneling. Tunn. Undergr. Space Technol. 21(2), 133–150 (2006) Kohestani, V. R., Bazarganlari, M. R., & Asgari Marnani, J. Prediction of maximum surface settlement caused by earth pressure balance shield tunneling using random forest. J. artif. intell. data min. 5(1), 127–135 (2017) Chen, R., Zhang, P., Wu, H., Wang, Z., & Zhong, Z. Prediction of shield tunneling-induced ground settlement using machine learning techniques. Front. Struct. Civ. Eng. 13, 1363–1378 (2019) Mahmoodzadeh, A. et al. Forecasting maximum surface settlement caused by urban tunneling. Autom. Constr. 120, 103375 (2020) Kim, D., Kwon, K., Pham, K., Oh, J. Y., & Choi, H. Surface settlement prediction for urban tunneling using machine learning algorithms with Bayesian optimization. Autom. Constr. 140, 104331 (2022a) Kim, D., Pham, K., Oh, J. Y., Lee, S. J., & Choi, H. Classification of surface settlement levels induced by TBM driving in urban areas using random forest with data-driven feature selection. Autom. Constr. 135, 104109 (2022b) Liu, L., Zhou, W., & Gutierrez, M. Effectiveness of predicting tunneling-induced ground settlements using machine learning methods with small datasets. J. Rock. Mech. Geotech. Eng. 14(4), 1028–1041 (2022) Jong, S. C., Ong, D. E. L., & Oh, E. State-of-the-art review of geotechnical-driven artificial intelligence techniques in underground soil-structure interaction. Tunn. Undergr. Space Technol. 113, 103946 (2021) Salimi, A., Faradonbeh, R. S., Monjezi, M., & Moormann, C. TBM performance estimation using a classification and regression tree (CART) technique. Bull. Eng. Geol. Environ. 77, 429–440 (2018) Breiman, L. Random forests. Mach. Learn. 45, 5–32 (2001) Chen, T., & Guestrin, C. Xgboost: A scalable tree boosting system. Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining , 785–794 (2016) Park, H., Oh, J. Y., Kim, D., & Chang, S. Monitoring and analysis of ground settlement induced by tunnelling with slurry pressure-balanced tunnel boring machine. Adv. Civ. Eng. (2018) Ahangari, K., Moeinossadat, S. R., & Behnia, D. Estimation of tunnelling-induced settlement by modern intelligent methods. Soils. Found. 55(4), 737–748 (2015) Ding, Z., Zhao, L. S., Zhou, W. H., & Bezuijen, A. Intelligent Prediction of Multi-Factor-Oriented Ground Settlement during TBM Tunneling in Soft Soil. Front. Built. Environ. 8, 848158 (2022) Fargnoli, V., Boldini, D., & Amorosi, A. TBM tunnelling-induced settlements in coarse-grained soils: The case of the new Milan underground line 5. Tunn. Undergr. Space Technol. 38, 336–347 (2013) Mooney, M. A., Grasmick, J., Kenneally, B., & Fang, Y. The role of slurry TBM parameters on ground deformation: Field results and computational modelling. Tunn. Undergr. Space Technol. 57, 257–264 (2016) Kwong, A. K. L., Ng, C. C. W., & Schwob, A. Control of settlement and volume loss induced by tunneling under recently reclaimed land. Undergr. Space 4(4), 289–301 (2019) Lee, H. K., Song, M. K., & Lee, S. S. Prediction of Subsidence during TBM Operation in Mixed-Face Ground Conditions from Realtime Monitoring Data. Appl. Sci. 11(24), 12130 (2021) Samadi, H., Hassanpour, J., & Farrokh, E. Maximum surface settlement prediction in EPB TBM tunneling using soft computing techniques. J. Phys: Conf. Ser. 1973(1) (2021) Liu, W. et al. A hybrid data-driven model for geotechnical reliability analysis. Reliab. Eng. Syst. Saf. 231, 108985 (2023) Sarna, S., Gutierrez, M., Mooney, M., & Zhu, M. Predicting upcoming collapse incidents during tunneling in rocks with continuation length based on influence zone. Rock. Mech. Rock. Eng. 55(10), 5905–5931 (2022) Su, J., Wang, Y., Niu, X., Sha, S., & Yu, J. Prediction of ground surface settlement by shield tunneling using XGBoost and Bayesian Optimization. Eng. Appl. Artif. Intell. 114, 105020 (2022) Zhou, J. et al. Estimation of the TBM advance rate under hard rock conditions using XGBoost and Bayesian optimization. Undergr. Space 6(5), 506–515 (2021) Zhang, Q., Hu, W., Liu, Z., & Tan, J. TBM performance prediction with Bayesian optimization and automated machine learning. Tunn. Undergr. Space Technol. 103, 103493 (2020) Sokolova, M., & Lapalme, G. A systematic analysis of performance measures for classification tasks. Inf. Process. Manage. 45(4): 427–437 (2009) Sun, Z. et al. A novel ensemble method for classifying imbalanced data. Pattern. Recognit. 48(5), 1623–1637 (2015) Zhou, Y., Li, S., Zhou, C., & Luo, H. Intelligent approach based on random forest for safety risk prediction of deep foundation pit in subway stations. J. Comput. Civ. Eng. 33(1), 05018004 (2019) Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4771476","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":339685734,"identity":"4e29ecc9-0fd1-4d81-98dc-ef182d2cd845","order_by":0,"name":"Kibeom Kwon","email":"","orcid":"","institution":"Korea University","correspondingAuthor":false,"prefix":"","firstName":"Kibeom","middleName":"","lastName":"Kwon","suffix":""},{"id":339685735,"identity":"247b9701-5cd7-4e67-8820-01d18facc624","order_by":1,"name":"Minkyu Kang","email":"","orcid":"","institution":"Korea Institute for Defense Analyses","correspondingAuthor":false,"prefix":"","firstName":"Minkyu","middleName":"","lastName":"Kang","suffix":""},{"id":339685736,"identity":"29d3036e-72e1-4441-841b-4e5c00fc8d80","order_by":2,"name":"Dongku Kim","email":"","orcid":"","institution":"Korea Institute of Civil Engineering and Building Technology (KICT)","correspondingAuthor":false,"prefix":"","firstName":"Dongku","middleName":"","lastName":"Kim","suffix":""},{"id":339685737,"identity":"f809d5b1-21a0-4ab7-a62c-e22d0d3c17e2","order_by":3,"name":"Khanh Pham","email":"","orcid":"","institution":"International University","correspondingAuthor":false,"prefix":"","firstName":"Khanh","middleName":"","lastName":"Pham","suffix":""},{"id":339685738,"identity":"f8a410b4-4f88-4d8a-8f65-b574c2dc2879","order_by":4,"name":"Hangseok Choi","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAx0lEQVRIiWNgGAWjYBACAwYeBoaEChsDuIgEcVrOpJGqhbHtMAlazNnPHnzwgO28Mb/04YMfGGrsGCRnH8CvxbInL9kggee2mWRfWrIEw7FkBmm+BAIOu8FjJpEgcdvG4AyPgQQD2wEGOR5CfgFrMTgH1ML/+QfDP6K1JBwwA9rCJsHYdoBBmqCWMznGBgkHko0le9jMLBL7knkkewhpOX7G8OHPf3aG/TzMj298+GYnJ3GGgBZUkMDAQMhZo2AUjIJRMAqIAQDrdznU2u5B6gAAAABJRU5ErkJggg==","orcid":"","institution":"Korea University","correspondingAuthor":true,"prefix":"","firstName":"Hangseok","middleName":"","lastName":"Choi","suffix":""}],"badges":[],"createdAt":"2024-07-20 06:06:00","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4771476/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4771476/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":62549305,"identity":"a848cf3e-4153-4a81-af3c-57e5e1c64943","added_by":"auto","created_at":"2024-08-15 16:48:51","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":13260,"visible":true,"origin":"","legend":"\u003cp\u003eLongitudinal geological profile of the addressed site.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-4771476/v1/b4f33ac522ba364db5277bca.png"},{"id":62549306,"identity":"29ec3699-8c2e-4b87-b62c-26be296a00a0","added_by":"auto","created_at":"2024-08-15 16:48:51","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":63773,"visible":true,"origin":"","legend":"\u003cp\u003ePearson’s coefficients of the features and the target.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-4771476/v1/829b988a0eafe994f6455dd9.png"},{"id":62549061,"identity":"63ba4734-d344-4ace-b198-87e488808b42","added_by":"auto","created_at":"2024-08-15 16:40:51","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":12404,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion matrix for the initial model.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-4771476/v1/0a595c0740acdf41b02d99d6.png"},{"id":62549797,"identity":"f1ed23d7-030c-4585-b3f3-f8c4e2b1b406","added_by":"auto","created_at":"2024-08-15 16:56:51","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":79433,"visible":true,"origin":"","legend":"\u003cp\u003eStructure of the DT-based initial model.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-4771476/v1/093a38ac07dcde656de968fd.png"},{"id":62549063,"identity":"6f97b9b0-40eb-4abe-9404-0364bb760caf","added_by":"auto","created_at":"2024-08-15 16:40:51","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":16524,"visible":true,"origin":"","legend":"\u003cp\u003eBox plots of each selected feature according to each class.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-4771476/v1/d5b827286205fb007ffe7038.png"},{"id":62550132,"identity":"a80db76b-dc0e-44d7-8412-91120726a451","added_by":"auto","created_at":"2024-08-15 17:04:51","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":15500,"visible":true,"origin":"","legend":"\u003cp\u003eIntelligent TBM operation support system developed in this study.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-4771476/v1/af97f0aeb30590a41ebda9b5.png"},{"id":62549310,"identity":"7b4d1ca8-9c2b-42aa-9323-3bc51292872e","added_by":"auto","created_at":"2024-08-15 16:48:51","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":5870,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion matrix for the developed support system.\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-4771476/v1/de08c803daced791bbdb1e8f.png"},{"id":62549066,"identity":"3bc5084d-eec6-481d-9ab2-ef256f627913","added_by":"auto","created_at":"2024-08-15 16:40:51","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":11383,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion matrices for the support system and comparison models.\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-4771476/v1/331bcff2252c6dd64417da15.png"},{"id":62549308,"identity":"d0084f44-4dda-4783-ad14-95b5fdf8f8a1","added_by":"auto","created_at":"2024-08-15 16:48:51","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":17803,"visible":true,"origin":"","legend":"\u003cp\u003eFlow chart of outlining the intelligent support system.\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-4771476/v1/c16583b12cc688737fee4d0a.png"},{"id":62549068,"identity":"0a1aced4-bca3-4d10-8596-ae194beef9b3","added_by":"auto","created_at":"2024-08-15 16:40:51","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":16583,"visible":true,"origin":"","legend":"\u003cp\u003eResults of error analysis.\u003c/p\u003e","description":"","filename":"10.png","url":"https://assets-eu.researchsquare.com/files/rs-4771476/v1/4d366822f92000d810ed5b99.png"},{"id":66042909,"identity":"c4aad9e7-527e-45f1-9fa7-f3e6404b8dfc","added_by":"auto","created_at":"2024-10-07 06:23:53","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1023412,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4771476/v1/bc0dbe77-00da-4842-aab8-88f7fff8d6bc.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Intelligent support system for ground settlement management during TBM tunneling by combining machine learning with statistical analysis","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe increasing population density and rising social costs associated with traffic congestion have driven a growing demand for tunnel construction [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Among various tunnel construction methods, tunnel boring machines (TBMs) have been extensively adopted owing to their superior eco-friendliness, stability, and constructability. However, the inherent geological, mechanical, and combined uncertainties encountered during TBM tunneling can result in adverse accidents, such as collapse and water/mud inflow [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Among these accidents, managing ground settlement within acceptable limits is crucial to prevent severe damage to existing surface and underground structures [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Therefore, it is essential to adjust TBM operation conditions based on a reliable support system which effectively predicts ground settlements.\u003c/p\u003e \u003cp\u003ePrevious studies on predicting ground settlement induced by tunneling have employed three primary approaches: empirical-statistical [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], analytical [\u003cspan additionalcitationids=\"CR8\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], and numerical [\u003cspan additionalcitationids=\"CR11\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e] methods. The empirical-statistical approach employs simple mathematical formulas based on prior tunnel excavation experience and data collected from extensive field observations. This method has been widely used in practical applications [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. However, applying this approach to TBM tunneling is challenging because it doesn't easily account for TBM operation features like thrust force and torque. The analytical approach provides closed-form solutions for predicting induced ground settlement through theoretical deductions. Despite its theoretical foundation, deriving suitable solutions using this approach is inherently complex due to necessary assumptions (e.g., isotropic, homogeneous, and incompressible soil) and uncertain relationships between settlement-inducing features, such as soil-structure interactions [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Furthermore, most analytical results cannot capture TBM operation features, similar to empirical results [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. With the rapid advancement of computer technology and numerical algorithms, several studies have utilized numerical approaches to predict ground settlement [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Unlike previous methods, numerical approaches can account for TBM operation and site-specific features (e.g., geology and geometry). However, they have limitations stemming from the use of constitutive models with substantial assumptions about uncertain material properties, operation features, and boundary conditions [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Moreover, considerable computational resources are needed to ensure the reliability of obtained results.\u003c/p\u003e \u003cp\u003eRecently, machine learning approaches have been introduced in various studies to predict TBM tunneling-induced ground settlement [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan additionalcitationids=\"CR21 CR22 CR23 CR24 CR25\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. This approach can consider TBM operation and site-specific features using extensive data and effectively analyze complex relationships between settlement-inducing features. However, most of machine learning-based studies have relied on black box-type algorithms, making it challenging to provide explicit predictive criteria. Furthermore, these algorithms may have limited practical applicability due to the need for extensive or localized feature information.\u003c/p\u003e \u003cp\u003eThis study aims to develop an intelligent TBM operation support system for ground settlement management. Correlation analysis is conducted to identify three key features utilized in the support system. The predictive criteria for three settlement classes are established through a combination of machine learning and statistical analysis. To validate the practical applicability of the developed system, a comparative analysis is conducted against ensemble machine learning models. Furthermore, error analysis is performed to evaluate the prediction performance of the support system regarding ground settlement values.\u003c/p\u003e"},{"header":"2. Overview of machine learning algorithms","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1. Decision tree (DT)\u003c/h2\u003e \u003cp\u003eDecision tree (DT) divides the predictor space into multiple regions, resulting in a tree-like structure developed based on a set of splitting rules [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. This structure comprises various nodes, including root, internal, and leaf nodes. A root node (containing the entire dataset) is split into two internal nodes (each containing a split dataset) based on an optimal feature that maximizes the information gain for a given index (e.g., the Gini index). Subsequently, the internal nodes are recursively divided until a leaf node is reached. In classification problems like this study, the final prediction is made using a set of these splitting rules. Unlike black box-type algorithms, DT is transparent as it explicitly provides the predictive criteria [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. DT was used to develop an initial model in this study.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2. Random forest (RF)\u003c/h2\u003e \u003cp\u003eRandom forest (RF) is a bagging-type ensemble machine learning algorithm introduced by [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] based on a set of decision trees. For each decision tree construction, a bootstrap sample (i.e., a randomly selected sample from the original training set with replacement) is drawn, with the number of data instances and features in the sample being randomly determined. The decision tree then evolves based on the selected bootstrap sample. In this manner, \u003cem\u003en\u003c/em\u003e bootstrap samples and their corresponding \u003cem\u003en\u003c/em\u003e decision trees are repeatedly constructed. By aggregating all the decision tree outputs, the final prediction is determined as the mode (in classification) or the average value (in regression). Because of its high prediction performance and low risk of overfitting, RF has been widely applied in various fields.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3. Extreme gradient boosting (XGB)\u003c/h2\u003e \u003cp\u003eExtreme gradient boosting (XGB) is a gradient boosting-type ensemble learning algorithm extensively applied in diverse domains, initially proposed by [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. The core of gradient boosting-type algorithms lies in the sequential creation of numerous weak learners, with each learner focused on rectifying the residual (i.e., the gradient of a loss function) left by its predecessor. The final strong learner emerges as an ensemble of these weak learners. Specifically, the predictions of these weak learners are then weighted based on their individual performance, and the aggregated prediction of these weighted predictions forms the outcome of the strong learner.\u003c/p\u003e \u003cp\u003eNevertheless, gradient boosting-type algorithms run the risk of constructing an excessively complex model by exclusively minimizing the residual, potentially leading to overfitting. To mitigate this risk, XGB incorporates an advanced objective function that combines the loss function with a regularization term, optimally balancing prediction performance and model complexity. In this study, RF and XGB were used to develop comparison models for validating the support system.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Dataset creation","content":"\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e3.1. Project overview\u003c/h2\u003e \u003cp\u003eThis study used a dataset obtained from a slurry shield TBM tunneling project in Hong Kong. Twin-bored slurry shield TBM subway tunnels with a total length of 850 m, comprising both up-track and down-track tunnels, were constructed at the tunneling site. The up-track tunnel was excavated five months after the down-track tunnel to minimize ground disturbance. These tunnels were categorized as shallow tunnels because their cover depth was less than twice the TBM excavation diameter [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. The detailed specifications of the slurry shield TBM are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSlurry shield TBM specifications.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDescription\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSpecification\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTBM excavation diameter\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7.4 m\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTBM length\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11.45 m\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMax. thrust force\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e47,897 kN\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMax. torque\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5 MN\u0026sdot;m\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOpening ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e29%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSegment diameter\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7.1 m (OD), 6.5 m (ID)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSegment width\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.5 m\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAfter the TBM passed through the second tunnel (i.e., up-track tunnel), daily settlement measurements were made at 253 locations along the tunnel alignment. The final settlement value at each location was determined as the average over five days, typically around the 20th day following the TBM passage. Four different ground types were encountered in the up-track alignment: fill, alluvium, completely decomposed granite (CDG), and highly decomposed granite (HDG). HDG and CDG can be classified as highly fractured normal to weak rock and very weak rock to weathered soil, respectively [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Representative geotechnical properties of these ground types are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The longitudinal geological profile along the tunnel alignment is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. More detailed information on data acquisition procedures and ground types can be referred to [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eGeotechnical properties of the encountered ground types [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGround type\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eN-value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCohesion [kPa]\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eFriction angle [\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:^\\circ\\:\\)\u003c/span\u003e\u003c/span\u003e]\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePermeability [m/s]\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFill\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e3.15\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\times\\:\\)\u003c/span\u003e\u003c/span\u003e10\u003csup\u003e-5\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlluvium\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eD\u003csup\u003e*\u003c/sup\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\u0026lt;\\)\u003c/span\u003e\u003c/span\u003e4: 10\u003c/p\u003e \u003cp\u003eD\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\ge\\:\\)\u003c/span\u003e\u003c/span\u003e4: 1.67 D\u0026thinsp;+\u0026thinsp;3.33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.32\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\times\\:\\)\u003c/span\u003e\u003c/span\u003e10\u003csup\u003e-5\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCDG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eD\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\u0026lt;\\)\u003c/span\u003e\u003c/span\u003e5: 15\u003c/p\u003e \u003cp\u003eD\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\ge\\:\\)\u003c/span\u003e\u003c/span\u003e5: 2.71 D\u0026thinsp;+\u0026thinsp;1.43\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e3.34\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\times\\:\\)\u003c/span\u003e\u003c/span\u003e10\u003csup\u003e-6\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHDG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e200\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.60\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\times\\:\\)\u003c/span\u003e\u003c/span\u003e10\u003csup\u003e-6\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003e\u003csup\u003e*\u003c/sup\u003eD: Depth [m]\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.2. Data pre-processing\u003c/h2\u003e \u003cp\u003eGenerally, features that induce settlement are categorized into three categories: geometry, geology, and TBM operation characteristics [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. In this study, the features commonly addressed and readily obtainable in various TBM tunneling projects were selected to ensure the practical applicability of the developed support system [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e].\u003c/p\u003e \u003cp\u003ePrevious studies considered the cover depth and the horizontal distance between the tunnel centerline and settlement monitoring points (HTM) as geometry features [\u003cspan additionalcitationids=\"CR24 CR25\" citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. These geometry features become particularly important for predicting settlement in shallow tunnels, similar to the site of this study. Geology features adopted here include the standard penetration test (SPT) N-value and the distance between the ground surface and the groundwater level (DSG). These features account for the potential influence of ground strength and groundwater on settlement. Notably, the SPT N-value, being more subdivided than other geotechnical properties (refer to Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e), is likely to influence machine learning predictions [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Concurrently, this study focused on five commonly measured TBM operation features in tunneling projects: thrust force, torque, face pressure, penetration rate, and grouting volume. These features have demonstrated their contributions to ground settlement in several previous studies [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan additionalcitationids=\"CR35 CR36 CR37\" citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eSince settlement measurements were conducted after the excavation of the up-track tunnel, this study used the features in each category, corresponding to the up-track tunnel excavation. Although, in certain chainages, settlements were measured at various locations with different horizontal distances. In this study, the settlements closest to the tunnel centerline were exclusively considered. Statistical descriptions of the settlement-inducing features and the settlements are presented in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eStatistical description of the selected features and settlements.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"9\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCategory\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFeature\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMin\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eQ1\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMedian\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eQ3\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMax\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eCOV\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eUnit\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eGeometry\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCover depth\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e6.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e7.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e8.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e12.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003em\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHTM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4.206\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e10.118\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e16.413\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e39.892\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003em\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eGeology\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eN-value\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e10.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e15.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e16.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e22.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e36.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDSG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.31\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e4.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e7.23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003em\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"4\" rowspan=\"5\"\u003e \u003cp\u003eOperation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThrust force\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e8,676\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e12,850\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e13,700\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e15,400\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e26,000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003ekN\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTorque\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.42\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eMN∙m\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFace pressure\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e2.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003ebar\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePenetration rate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e13.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e27.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e35.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e40.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e48.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003emm/min\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrouting volume\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e6.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6.43\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e6.60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e7.60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003em\u003csup\u003e3\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSettlement\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-6.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e8.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e16.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.63\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003emm\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003csup\u003e*\u003c/sup\u003e COV is the coefficient of variation, denoted as the standard deviation normalized by the mean.\u003c/p\u003e \u003cp\u003eThis study categorized the measured settlements (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:S\\)\u003c/span\u003e\u003c/span\u003e) into three classes based on their magnitude: heaving (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:S\u0026lt;0\\:mm\\)\u003c/span\u003e\u003c/span\u003e), normal (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:0\\:mm\u0026lt;S\\le\\:10\\:mm\\)\u003c/span\u003e\u003c/span\u003e), and large settlement (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:S\u0026gt;10\\:mm\\)\u003c/span\u003e\u003c/span\u003e), as summarized in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. A total of 129 instances comprises nine features (including geometry, geology, and operation features) and one target variable (settlement class).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCategorized settlement classes defined in this study.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eClass\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCount\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eProportion\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHeaving (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\varvec{S}\u0026lt;0\\:\\varvec{m}\\varvec{m}\\)\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e12.4%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNormal (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:0\\:\\varvec{m}\\varvec{m}\u0026lt;\\varvec{S}\\le\\:10\\:\\varvec{m}\\varvec{m}\\)\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e71.3%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLarge settlement (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\varvec{S}\u0026gt;10\\:\\varvec{m}\\varvec{m}\\)\u003c/span\u003e\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e16.3%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.3. Data analysis\u003c/h2\u003e \u003cp\u003eIdentifying highly correlated features with the categorized settlement class among all the collected features is important. To simplify the developed support system for practical use by TBM operators, only a small number of features were selectively adopted. Through correlation analysis, the relationships between these features and the target variable (i.e., the settlement class) were identified. The results of this correlation analysis, presented by Pearson\u0026rsquo;s coefficient, are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe correlation analysis indicated that DSG, thrust force, torque, and face pressure were the most relevant features with respect to the settlement class. Specially, thrust force and torque exhibited strong negative correlations with the settlement class, represented by Pearson\u0026rsquo;s coefficients of -0.49 and \u0026minus;\u0026thinsp;0.48, respectively. In simpler terms, increases and decreases in these features (i.e., thrust force and torque) were closely associated with the heaving and large settlement classes, respectively. This observation aligns with findings from previous studies [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]. Consequently, thrust force and torque were selected as the key features for the support system.\u003c/p\u003e \u003cp\u003eFace pressure has been a critical parameter for controlling induced ground settlements due to its significant influence on face stability during TBM tunneling [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e reinforces this perspective by demonstrating a strong correlation between face pressure and the settlement class, indicated by a Pearson\u0026rsquo;s coefficient of -0.44. Nevertheless, it is important to note that face pressure also exhibited a strong correlation with thrust force (i.e., Pearson\u0026rsquo;s coefficient\u0026thinsp;\u0026ge;\u0026thinsp;0.7) and had a low coefficient of variation (COV), as shown in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. Similarly, despite DSG showing a high Pearson\u0026rsquo;s coefficient of -0.50, it was also strongly correlated with thrust force. Moreover, recording DSG can be relatively uncertain since it was interpolated from results derived from various exploration boreholes. Consequently, to enhance the practical applicability of the support system, this study replaced these two features, face pressure and DSG, with thrust force.\u003c/p\u003e \u003cp\u003eIn the support system development, most of the features from Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, which demonstrated weak correlations with the settlement class, were excluded. However, even though the correlation between N-value and the settlement class was low, this study retained N-value as a geology feature. This decision was made to account for the significance of weak ground, which typically corresponds to a high risk of ground settlement [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. As a result, the dataset was streamlined to consist of three features (i.e., thrust force, torque, and N-value) along with the target variable (i.e., the settlement class).\u003c/p\u003e \u003c/div\u003e"},{"header":"4. System development","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n \u003ch2\u003e4.1. Criterion for heaving class\u003c/h2\u003e\n \u003cp\u003eThe primary objective of this study is to develop an intelligent TBM operation support system with ground settlement prediction induced by TBM tunneling. This system classifies settlement into three classes, as outlined in Table \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e, using three key features: thrust force, torque, and N-value. However, establishing predictive criteria for this system from the beginning, especially when dealing with four-dimensional data encompassing thrust force, torque, N-value, and settlement class, is a complex challenge. Therefore, this study first constructed an initial model employing a machine learning approach to provide preliminary predictive criteria.\u003c/p\u003e\n \u003cp\u003eThe initial model was produced using the DT algorithm, chosen for its capacity to offer clear visualization and an intuitive interpretation of predictive criteria, in contrast to black box-type algorithms such as the RF and XGB algorithms. Subsequently, the dataset was divided into training and test sets, with 80% of the instances allocated for training and the remaining 20% for testing. It is important to note that the same test set was used for evaluating prediction accuracies corresponding to the initial model, the support system, and comparison models. In addition, this study utilized Bayesian optimization (BO) combined with 5-fold cross-validation for hyperparameter tuning.\u003c/p\u003e\n \u003cp\u003eBO is an efficient strategy for finding optimal solutions, such as hyperparameter combinations that maximize an unknown objective function (e.g., accuracy and coefficient of determination). It relies on two core components: surrogate model and acquisition function. The surrogate model estimates the posterior distribution of the objective function using prior knowledge derived from previously evaluated samples (e.g., hyperparameter combinations) based on Bayes\u0026rsquo; theorem [\u003cspan class=\"CitationRef\"\u003e41\u003c/span\u003e]. The acquisition function then leverages this estimated distribution to determine the next sample for evaluation [\u003cspan class=\"CitationRef\"\u003e42\u003c/span\u003e]. In this study, a Gaussian process (GP) was employed as the surrogate model, while expected improvement (EI) were adopted as the acquisition function. Detailed information on BO based on GP and EI can be found in [\u003cspan class=\"CitationRef\"\u003e43\u003c/span\u003e].\u003c/p\u003e\n \u003cp\u003eTable \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e presents the explored hyperparameters and their corresponding search spaces. It can be noted that the\u0026nbsp;\u003cem\u003emax_depth\u003c/em\u003e hyperparameter was set to less than five to mitigate overfitting and reduce complexity.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u0026nbsp;\u003ctable id=\"Tab5\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eSearch spaces for the explored DT hyperparameters.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eHyperparameter\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eDescription\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSearch space\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003emax_depth\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMaximum depth of tree model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2\u0026ndash;4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003emin_samples_split\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMinimum number of samples required to split a node\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3\u0026ndash;10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003emin_samples_leaf\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMinimum number of samples required in a leaf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3\u0026ndash;10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003eThe initial model was developed with the following optimal hyperparameters: \u003cem\u003emax_depth\u003c/em\u003e\u0026thinsp;=\u0026thinsp;4, \u003cem\u003emin_samples_split\u003c/em\u003e\u0026thinsp;=\u0026thinsp;10, and \u003cem\u003emin_samples_leaf\u003c/em\u003e\u0026thinsp;=\u0026thinsp;3. Table \u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003e and Fig. 3 show the prediction performance in the training and test phases. In this study, four performance metrics were used: accuracy, F1 score, precision, and recall. Macro-averaging precision and recall were adopted, where macro-averaging assigns equal weight to each class, regardless of the number of instances in each class [\u003cspan class=\"CitationRef\"\u003e44\u003c/span\u003e].\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u0026nbsp;\u003ctable id=\"Tab6\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003ePrediction performance for each class obtained by the initial model.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePerformance metric\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTraining\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eTest\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.825\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.808\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eF1 score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.751\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.732\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.858\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.783\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRecall\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.668\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.687\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003eAlthough over two-thirds of the heaving class in both training and test sets were correctly predicted, significant prediction errors were observed in the test phase for the large settlement class, with the recall for that class of 0.500. Throughout the training and test phases, only eight out of the 21 cases for the large settlement class were accurately predicted, while the rest were misclassified as the normal class. This misclassification issue can be attributed to the initial model\u0026rsquo;s lack of emphasis on predicting the large settlement class, which is a common challenge when applying machine learning algorithms to imbalanced datasets.\u003c/p\u003e\n \u003cp\u003eConsequently, according to the DT structure of the initial model (Fig. \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e), the predictive criterion for the heaving class (Eq.\u0026nbsp;(\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e)) was adopted for the support system developed in this study. However, additional analyses were required to address the misclassifications between the normal and large settlement classes, thereby deriving their predictive criteria.\u003c/p\u003e\n \u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e$$\\:N\\_value\\le\\:28\\:and\\:Thrust\\:force\u0026gt;\\text{17,695}\\:kN$$\u003c/div\u003e\n \u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\n \u003ch2\u003e4.2. Criterion for large settlement class\u003c/h2\u003e\n \u003cp\u003eIn this study, statistical analysis was employed to establish the criteria for predicting the normal and large settlement classes. Typically, in the statistical analysis of dataset distributions, five key indices are considered: the minimum and maximum values, the lower and upper quartiles (Q1 and Q3), and the median. These indices of three features (thrust force, torque, and N-value) are summarized for each settlement class in Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e and visually depicted in a box plot shown in Fig. 5.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable id=\"Tab7\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 7\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eStatistical description of the selected features corresponding to each class.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eIndices\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003eThrust force [kN]\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003eTorque [MN∙m]\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003eN-value\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eHeaving\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eNormal\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eLarge\u003c/p\u003e\n \u003cp\u003esettlement\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eHeaving\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eNormal\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eLarge\u003c/p\u003e\n \u003cp\u003esettlement\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eHeaving\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eNormal\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eLarge\u003c/p\u003e\n \u003cp\u003esettlement\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e14,500\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e9,628\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e8,822\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eQ1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e17,890\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e12,700\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e12,000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.325\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e15.25\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMedian\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e19,500\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e13,500\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e13,100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e17.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eQ3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e20,600\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e14,800\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e14,444\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.775\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e24.75\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMax\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e25,800\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e26,000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e15,500\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e31\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003eAccording to Fig.\u0026nbsp;5, it becomes evident that both thrust force and torque decrease as settlement magnitude increases, following the order of the heaving, normal, and large settlement classes. While the decline in thrust force is less pronounced, the trend in torque is more noticeable. Hence, this study defined the criterion for predicting the large settlement class as follows: instances that do not meet the heaving class criterion while concurrently exhibiting torque values below the lower quartile (Q1) for the normal class. Based on Eq.\u0026nbsp;(\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e) and Table \u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e, the criterion for predicting the large settlement class is summarized in Eq.\u0026nbsp;(\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e\n \u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e$$\\:[N\\_value\u0026gt;28\\:or\\:Thrust\\:force\\le\\:\\text{17,695}\\:kN]\\:and\\:[Torque\u0026lt;0.5\\:MN\\cdot\\:m]$$\u003c/div\u003e\n \u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n \u003ch2\u003e4.3. Criterion for normal class\u003c/h2\u003e\n \u003cp\u003eThe criterion for predicting the normal class can be determined as follows: instances that do not meet the conditions specified in equations (\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e) and (\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). Therefore, a data-driven support system for predicting ground settlement, comprising three criteria for predicting each settlement class, was formulated by combining machine learning with statistical analysis. Figure \u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003e illustrates the predictive criteria within the developed support system.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"5. System application","content":"\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\n \u003ch2\u003e5.1. Prediction performance\u003c/h2\u003e\n \u003cp\u003eThe prediction performances of the developed support system were assessed using the test set, comprising a total of 26 instances, as described in Section \u003cspan class=\"InternalRef\"\u003e3.2\u003c/span\u003e. Table \u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e presents the prediction performances of both the initial model and the support system when evaluated on the same test set, while Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e illustrates a confusion matrix for the developed support system.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable id=\"Tab8\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 8\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003ePrediction performance of the initial model and the developed support system.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"3\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePerformance metric\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eInitial model\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSupport system\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.808\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.847\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eF1 score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.732\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.784\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.783\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.798\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRecall\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.687\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.770\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003eAccording to Table \u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e, the support system enhanced prediction performance compared to the initial model, achieving an accuracy of 0.847, an F1 score of 0.784, a precision of 0.798, and a recall of 0.770. Notably, in the large settlement class, the support system correctly predicted three out of four instances, whereas the initial model correctly predicted only two. Meanwhile, the number of correctly predicted instances for the heaving and normal classes remained consistent. Importantly, all incorrectly predicted instances were classified into adjacent classes, reducing the risk of applying conflicting countermeasures. Consequently, the support system, which integrates machine learning with statistical analysis, outperformed the DT-based initial model, while both systems can explicitly provide predictive criteria.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\n \u003ch2\u003e5.2. Comparative analysis\u003c/h2\u003e\n \u003cp\u003eThe purpose of this section is to validate the practical applicability of the developed support system through a comparative analysis with different comparison models. Typically, bagging and boosting-based ensemble machine learning algorithms have been widely employed to address imbalanced datasets [\u003cspan class=\"CitationRef\"\u003e45\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e46\u003c/span\u003e]. As shown in Table \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e, a substantial proportion of the dataset corresponds to the normal class, signifying its imbalance. Consequently, this study developed two comparison models using representative bagging and boosting-based ensemble machine learning algorithms, namely RF and XGB. It can be noted that all the features listed in Table \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e were used in building the comparison models to highlight the simplicity of the developed support system that employs only three features. For hyperparameter tuning of these comparison models, BO and 5-fold cross-validation were employed. The search spaces for RF and XGB hyperparameters are described in Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003e.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable id=\"Tab9\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 9\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eSearch spaces of RF and XGB hyperparameters.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"3\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAlgorithm\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eDescription\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSearch space\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"4\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003en_estimators\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e100\u0026ndash;1000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003emax_depth\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2\u0026ndash;30\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003emin_samples_split\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3\u0026ndash;10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003emin_samples_leaf\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3\u0026ndash;10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"6\"\u003e\n \u003cp\u003eXGB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003en_estimators\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e100\u0026ndash;1000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003emax_depth\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2\u0026ndash;30\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003emin_child_weight\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u0026ndash;5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003egamma\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u0026ndash;1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003esubsample\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u0026ndash;1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003elearning-rate\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.01\u0026ndash;1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003eTable\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e10\u003c/span\u003e presents the prediction performances of the developed support system and the comparison models (i.e., RF and XGB models) when evaluated on the same test set, while Fig. 8 illustrates the confusion matrices for each model.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable id=\"Tab10\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 10\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003ePrediction performance of the support system and comparison models.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"4\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePerformance metric\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSupport system\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eRF model\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eXGB model\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.847\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.808\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.769\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eF1 score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.784\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.715\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.649\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.798\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.680\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.630\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRecall\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.770\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.753\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.670\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003eThe comparative analysis demonstrated that all the performance metrics of the support system surpassed those of the RF and XGB models. This suggests that the support system, which combines machine learning with statistical analysis, is more effective in predicting ground settlement than the ensemble machine learning algorithms. Moreover, the practical applicability of the support system was validated, as it was developed using only three features (i.e., thrust force, torque, and N-value) and provided explicit predictive criteria, as proposed in this study. The overall flow chart outlining the intelligent support system developed in this study is presented in Fig. \u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003e.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\n \u003ch2\u003e5.3. Error analysis for the support system\u003c/h2\u003e\n \u003cp\u003eThis study performed error analysis examining the prediction results for the instances within the test set, with the consideration of their settlement values. Figure \u003cspan class=\"InternalRef\"\u003e10\u003c/span\u003e displays the settlement values of these instances and indicates whether the developed support system predicted them correctly or incorrectly. As shown in Fig. \u003cspan class=\"InternalRef\"\u003e10\u003c/span\u003e, it was evident that three out of four incorrectly predicted instances had settlement values within \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\pm\\:\\)\u003c/span\u003e\u003c/span\u003e3 mm. These inaccuracies, closely associated with subtle ground deformations, were expected to have a relatively minor impact on the safety and efficiency of TBM tunneling projects. In contrast, it should be noted that the instance characterized by a substantial settlement value of 16 mm was incorrectly predicted, which could have significant adverse consequences. As shown in Table \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e, this instance represented the maximum settlement value in the entire dataset. It can be inferred that the developed support system faced challenges in accurately predicting this instance due to the scarcity of large settlement data, particularly those exceeding 15 mm. This observation highlights the need for an extensive dataset encompassing a substantial number of instances with large settlement values. Such a dataset would not only enhance the prediction accuracy of the developed support system but also ensure the safety and efficiency of TBM tunneling projects.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\n \u003ch2\u003e5.4. Limitation of this study\u003c/h2\u003e\n \u003cp\u003eDespite achieving the support system with superior prediction performances and practical applicability, this study has a limitation. The support system was constructed using a relatively small dataset of only 129 instances obtained from a single specific tunneling site, potentially limiting its generalization capability. Therefore, future studies could focus on building an integrated database from multiple TBM tunneling projects with similar working conditions to enhance and complement the support system\u0026rsquo;s performance.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"6. Conclusions","content":"\u003cp\u003eThis study introduced an intelligent TBM operation support system for ground settlement management, which combined machine learning with statistical analysis. The practical applicability of the support system was validated by comparison with ensemble machine learning models. The principal findings of this study are summarized as follows.\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThe support system incorporated three predictive criteria employing three key features: thrust force, torque, and N-value. First, a DT-based initial model was employed to establish the predictive criterion for heaving. Subsequently, statistical analysis was used to determine the predictive criterion for large settlement. Finally, the predictive criterion for the normal class involved instances that did not meet the preceding criteria.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThe support system, employing just three key features, outperformed RF and XGB models that used nine features. Moreover, the support system provides explicit predictive criteria, unlike the comparison models. These advantages underscore the practical applicability of the developed support system.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eOut of the four instances incorrectly predicted by the support system, three were associated with settlement values within \u0026plusmn;\u0026thinsp;3 mm, which generally have a minor impact on TBM tunneling. Conversely, the remained one instance had a substantial settlement value of 16 mm, indicating the need for support system improvement by incorporating a more extensive dataset, particularly encompassing significantly large settlement data.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData availability \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and/or analyzed during the current study are not publicly available due to\u0026nbsp;site information security issues\u0026nbsp;but are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research was supported by the National R\u0026amp;D Project for Smart Construction Technology (RS-2020-KA157074) and for Consecutive Excavation Technological Development Project of Tunnel Boring Machine (RS-2022-00144188) funded by the Korea Agency for Infrastructure Technology Advancement under the Ministry of Land, Infrastructure, and Transport.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eContributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eKibeom Kwon:\u003c/strong\u003e Conceptualization, Methodology, Writing - Original Draft. \u003cstrong\u003eMinkyu Kang:\u003c/strong\u003e Data Curation, Formal analysis, Validation. \u003cstrong\u003eDongku Kim:\u003c/strong\u003e Data Curation, Software, Visualization. \u003cstrong\u003eKhanh Pham:\u003c/strong\u003e Software, Formal analysis, Methodology. \u003cstrong\u003eHangseok Choi:\u003c/strong\u003e Conceptualization, Supervision, Writing - Review \u0026amp; Editing.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eKwon, K, Kang, M, Kim, D, \u0026amp; Choi, H. Prioritization of hazardous zones using an advanced risk management model combining the analytic hierarchy process and fuzzy set theory. Sustainability 15(15), 12018 (2023)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSousa RL, Einstein HH. Lessons from accidents during tunnel construction. Tunn. Undergr. Space Technol. 113, 103916 (2021)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eErcelebi, S. G, \u0026Ccedil;opur, H., \u0026amp; Ocak, I. Surface settlement predictions for Istanbul Metro tunnels excavated by EPB-TBM. Environ. Earth. Sci. 62, 357\u0026ndash;365 (2011)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChung, H. et al. A causal network-based risk matrix model applicable to shield TBM tunneling projects. Sustainability 13(9), 4846 (2021)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePeck, R. B. Deep excavations and tunnelling in soft ground. Proceedings of the 7th International Conference on Soil Mechanics and Foundation Engineering, Mexico City, Mexico, 225\u0026ndash;290 (1969)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAtkinson, J. H. \u0026amp; Potts, D. M. Subsidence above shallow tunnels in soft ground. J. Geotech. Engrg. Div. 103(4), 307\u0026ndash;325 (1977)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSagaseta, C. Analysis of undrained soil deformation due to ground loss. Geotechnique 37(3), 301\u0026ndash;320 (1987)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLoganathan, N., \u0026amp; Poulos, H. G. Analytical prediction for tunneling-induced ground movements in clays. J. Geotech. Geoenviron. Eng. 124(9), 846\u0026ndash;856 (1998)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVerruijt, A. A complex variable solution for a deforming circular tunnel in an elastic half-plane. Int. J. Numer. Anal. Methods Geomech. 21(2), 77\u0026ndash;89 (1997)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFranzius, J. N., Potts, D. M., \u0026amp; Burland, J. B. The influence of soil anisotropy and K0 on ground surface movements resulting from tunnel excavation. Geotechnique 55(3), 189\u0026ndash;199 (2005)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChakeri, H., Ozcelik, Y., \u0026amp; Unver, B. Effects of important factors on surface settlement prediction for metro tunnel excavated by EPB. Tunn. Undergr. Space Technol. 36, 14\u0026ndash;23 (2013)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang, H. et al. Simplified procedure for finite element analysis of the longitudinal performance of shield tunnels considering spatial soil variability in longitudinal direction. Comput. Geotech. 64, 132\u0026ndash;145 (2015)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoghaddasi, M. R., \u0026amp; Noorian-Bidgoli, M. ICA-ANN, ANN and multiple regression models for prediction of surface settlement caused by tunneling. Tunn. Undergr. Space Technol. 79, 197\u0026ndash;209 (2018)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou, J. et al. Feasibility of random-forest approach for prediction of ground settlements induced by the construction of a shield-driven tunnel. Int. J. Geomech. 17(6): 04016129 (2017)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, D., Pham, K., Park, S., Oh, J. Y., \u0026amp; Choi, H. Determination of effective parameters on surface settlement during shield TBM. Geomech. Eng. 21(2), 153\u0026ndash;164 (2020)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, R. P., Song, X., Meng, F. Y., Wu, H. N., \u0026amp; Lin, X. T. Analytical approach to predict tunneling-induced subsurface settlement in sand considering soil arching effect. Comput. Geotech. 141, 104492 (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K., Oh, J. Y., Lee, H., Kim, D., \u0026amp; Choi, H. Critical face pressure and backfill pressure in shield TBM tunneling on soft ground. Geomech. Eng. 15(3), 823\u0026ndash;831 (2018)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou, X., Zhao, C., \u0026amp; Bian, X. Prediction of maximum ground surface settlement induced by shield tunneling using XGBoost algorithm with golden-sine seagull optimization. Comput. Geotech. 154, 105156 (2023)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYe, X. W., Jin, T., \u0026amp; Chen, Y. M. Machine learning-based forecasting of soil settlement induced by shield tunneling construction. Tunn. Undergr. Space Technol. 124, 104452 (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSuwansawat, S., \u0026amp; Einstein, H. H. Artificial neural networks for predicting the maximum surface settlement caused by EPB shield tunneling. Tunn. Undergr. Space Technol. 21(2), 133\u0026ndash;150 (2006)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKohestani, V. R., Bazarganlari, M. R., \u0026amp; Asgari Marnani, J. Prediction of maximum surface settlement caused by earth pressure balance shield tunneling using random forest. J. artif. intell. data min. 5(1), 127\u0026ndash;135 (2017)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, R., Zhang, P., Wu, H., Wang, Z., \u0026amp; Zhong, Z. Prediction of shield tunneling-induced ground settlement using machine learning techniques. Front. Struct. Civ. Eng. 13, 1363\u0026ndash;1378 (2019)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMahmoodzadeh, A. et al. Forecasting maximum surface settlement caused by urban tunneling. Autom. Constr. 120, 103375 (2020)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, D., Kwon, K., Pham, K., Oh, J. Y., \u0026amp; Choi, H. Surface settlement prediction for urban tunneling using machine learning algorithms with Bayesian optimization. Autom. Constr. 140, 104331 (2022a)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, D., Pham, K., Oh, J. Y., Lee, S. J., \u0026amp; Choi, H. Classification of surface settlement levels induced by TBM driving in urban areas using random forest with data-driven feature selection. Autom. Constr. 135, 104109 (2022b)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, L., Zhou, W., \u0026amp; Gutierrez, M. Effectiveness of predicting tunneling-induced ground settlements using machine learning methods with small datasets. J. Rock. Mech. Geotech. Eng. 14(4), 1028\u0026ndash;1041 (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJong, S. C., Ong, D. E. L., \u0026amp; Oh, E. State-of-the-art review of geotechnical-driven artificial intelligence techniques in underground soil-structure interaction. Tunn. Undergr. Space Technol. 113, 103946 (2021)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSalimi, A., Faradonbeh, R. S., Monjezi, M., \u0026amp; Moormann, C. TBM performance estimation using a classification and regression tree (CART) technique. Bull. Eng. Geol. Environ. 77, 429\u0026ndash;440 (2018)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBreiman, L. Random forests. Mach. Learn. 45, 5\u0026ndash;32 (2001)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, T., \u0026amp; Guestrin, C. Xgboost: A scalable tree boosting system. \u003cem\u003eProceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining\u003c/em\u003e, 785\u0026ndash;794 (2016)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePark, H., Oh, J. Y., Kim, D., \u0026amp; Chang, S. Monitoring and analysis of ground settlement induced by tunnelling with slurry pressure-balanced tunnel boring machine. Adv. Civ. Eng. (2018)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAhangari, K., Moeinossadat, S. R., \u0026amp; Behnia, D. Estimation of tunnelling-induced settlement by modern intelligent methods. Soils. Found. 55(4), 737\u0026ndash;748 (2015)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDing, Z., Zhao, L. S., Zhou, W. H., \u0026amp; Bezuijen, A. Intelligent Prediction of Multi-Factor-Oriented Ground Settlement during TBM Tunneling in Soft Soil. Front. Built. Environ. 8, 848158 (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFargnoli, V., Boldini, D., \u0026amp; Amorosi, A. TBM tunnelling-induced settlements in coarse-grained soils: The case of the new Milan underground line 5. Tunn. Undergr. Space Technol. 38, 336\u0026ndash;347 (2013)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMooney, M. A., Grasmick, J., Kenneally, B., \u0026amp; Fang, Y. The role of slurry TBM parameters on ground deformation: Field results and computational modelling. Tunn. Undergr. Space Technol. 57, 257\u0026ndash;264 (2016)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKwong, A. K. L., Ng, C. C. W., \u0026amp; Schwob, A. Control of settlement and volume loss induced by tunneling under recently reclaimed land. Undergr. Space 4(4), 289\u0026ndash;301 (2019)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee, H. K., Song, M. K., \u0026amp; Lee, S. S. Prediction of Subsidence during TBM Operation in Mixed-Face Ground Conditions from Realtime Monitoring Data. Appl. Sci. 11(24), 12130 (2021)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSamadi, H., Hassanpour, J., \u0026amp; Farrokh, E. Maximum surface settlement prediction in EPB TBM tunneling using soft computing techniques. J. Phys: Conf. Ser. 1973(1) (2021)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, W. et al. A hybrid data-driven model for geotechnical reliability analysis. Reliab. Eng. Syst. Saf. 231, 108985 (2023)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSarna, S., Gutierrez, M., Mooney, M., \u0026amp; Zhu, M. Predicting upcoming collapse incidents during tunneling in rocks with continuation length based on influence zone. Rock. Mech. Rock. Eng. 55(10), 5905\u0026ndash;5931 (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSu, J., Wang, Y., Niu, X., Sha, S., \u0026amp; Yu, J. Prediction of ground surface settlement by shield tunneling using XGBoost and Bayesian Optimization. Eng. Appl. Artif. Intell. 114, 105020 (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou, J. et al. Estimation of the TBM advance rate under hard rock conditions using XGBoost and Bayesian optimization. Undergr. Space 6(5), 506\u0026ndash;515 (2021)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, Q., Hu, W., Liu, Z., \u0026amp; Tan, J. TBM performance prediction with Bayesian optimization and automated machine learning. Tunn. Undergr. Space Technol. 103, 103493 (2020)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSokolova, M., \u0026amp; Lapalme, G. A systematic analysis of performance measures for classification tasks. Inf. Process. Manage. 45(4): 427\u0026ndash;437 (2009)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun, Z. et al. A novel ensemble method for classifying imbalanced data. Pattern. Recognit. 48(5), 1623\u0026ndash;1637 (2015)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou, Y., Li, S., Zhou, C., \u0026amp; Luo, H. Intelligent approach based on random forest for safety risk prediction of deep foundation pit in subway stations. J. Comput. Civ. Eng. 33(1), 05018004 (2019)\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Intelligent support system, Tunnel boring machine, Ground settlement, Machine learning, Statistical analysis","lastPublishedDoi":"10.21203/rs.3.rs-4771476/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4771476/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eGround settlement management is crucial in tunnel boring machine (TBM) tunneling. Previous studies on predicting ground settlement have required substantial assumptions or information, making it challenging to explicitly determine their predictive criteria. This study developed an intelligent TBM operation support system for ground settlement management, by combining machine learning and statistical analysis. Initially, measured settlements were categorized into three classes: heaving, normal, and large settlement. Based on three key features selected through correlation analysis, the predictive criterion for the heaving class was determined using an initial model based on a decision tree algorithm. Subsequently, through statistical analysis, the predictive criteria for the normal and large settlement classes were established. The developed support system, using only three key features, achieved an accuracy of 0.847, F1 score of 0.784, precision of 0.798, and recall of 0.770, outperforming two ensemble machine learning models that used nine features. Moreover, the system can provide explicit predictive criteria, enhancing its practical applicability. Error analysis revealed that among the four instances misclassified by the support system, three pertained to minor settlements within \u0026plusmn;\u0026thinsp;3 mm.\u003c/p\u003e","manuscriptTitle":"Intelligent support system for ground settlement management during TBM tunneling by combining machine learning with statistical analysis","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-08-15 16:40:46","doi":"10.21203/rs.3.rs-4771476/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"d4024c66-050d-4424-9767-e954ebd51bd2","owner":[],"postedDate":"August 15th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":35986581,"name":"Physical sciences/Engineering/Civil engineering"},{"id":35986582,"name":"Physical sciences/Mathematics and computing/Scientific data"}],"tags":[],"updatedAt":"2024-10-07T06:23:24+00:00","versionOfRecord":[],"versionCreatedAt":"2024-08-15 16:40:46","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4771476","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4771476","identity":"rs-4771476","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}