Acta Geotechnica (2022) 17:1343–1366

https://doi.org/10.1007/s11440-022-01450-7 (0123456789().,-volV)(0123456789().
,- volV)

RESEARCH PAPER

Predicting tunnel squeezing using support vector machine optimized

by whale optimization algorithm
Jian Zhou1 • Shuangli Zhu1 • Yingui Qiu1 • Danial Jahed Armaghani2 • Annan Zhou3 • Weixun Yong1

Received: 21 September 2021 / Accepted: 4 January 2022 / Published online: 29 January 2022
Ó The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2022

Abstract
The squeezing behavior of surrounding rock can be described as the time-dependent large deformation during tunnel
excavation, which appears in special geological conditions, such as weak rock masses and high in situ stress. Several
problems such as budget increase and construction period extension can be caused by squeezing in rock mass. It is
significant to propose a model for accurate prediction of rock squeezing. In this research, the support vector machine
(SVM) as a machine learning model was optimized by the whale optimization algorithm (WOA), WOA-SVM, to classify
the tunnel squeezing based on 114 real cases. The role of WOA in this system is to optimize the hyper-parameters of SVM
model for receiving a higher level of accuracy. In the established database, five input parameters, i.e., buried depth, support
stiffness, rock tunneling quality index, diameter and the percentage strain, were used. In the process of model classification,
different effective parameters of SVM and WOA were considered, and the optimum parameters were designed. To
examine the accuracy of the WOA-SVM, the base SVM, ANN (refers to the multilayer perceptron) and GP (refers to the
Gaussian process classification) were also constructed. Evaluation of these models showed that the optimized WOA-SVM
is the best model among all proposed models in classifying the tunnel squeezing. It has the highest accuracy (approxi-
mately 0.9565) than other un-optimized individual classifiers (SVM, ANN, and GP). This was obtained based on results of
different performance indexes. In addition, according to sensitivity analysis, the percentage strain is highly sensitive to the
model, followed by buried depth and support stiffness. That means, e, H and K are the best combination of parameters for
the WOA–SVM model.

Keywords Classification modeling  Squeezing  SVM  Tunnel  WOA algorithm

1 Introduction surrounding rock and various forms of supporting structure

failures [19]. There are objective and subjective factors for
Tunnel squeezing refers to the occurrence of large amount the occurrence of tunnel squeezing, where the objective
of deformation in surrounding rock mass rock, which is conditions involve rock properties, tectonic stress, tunnel
normally more than the designed deformation. This phe- dimensions, rock type, high in situ stress and large radius
nomenon, which takes a long time to form, causes many or span [8]. On the other hand, the typical subjective fac-
difficulties during and after construction of tunnels tors are associated with support installation, in which the
[6, 11, 69]. The squeezing behavior of surrounding rock deformation can be restrained if the support is installed on
can be described as the time-dependent large deformation time [27, 42, 62]. Tunnel squeezing may cause several
during tunnel excavation, which is essentially related to unwanted issues, e.g., budget increase, construction period
creep created by exceeding the ultimate shear stress extension and construction safety [8, 22]. In order to
[8, 19, 25, 59, 69]. Different studies showed that the overcome these issues, many attempts have been done by
compressive surrounding rock has the deformation features various scholars, and they suggested several approaches for
of large deformation amount, long deformation duration, predicting tunnel squeezing, including empirical, semiem-
high deformation speed, large destruction range of pirical and theoretical methods [4, 24, 32, 33, 61, 68]. With
the development of the computer science and various
available technologies, numerical simulation and classical
Extended author information available on the last page of the article

123
1344 Acta Geotechnica (2022) 17:1343–1366

statistics methods have been widely used in tunnel will be squeezed. So far, most of the existing forecasting
squeezing prediction [13, 23, 37–39, 65]. methods can be used to distinguish between squeezing and
In recent years, the successful applications of machine non-squeezing. However this article refers to the multi-
learning (ML) methods in solving regression, classification class SVM proposed by Sun et al. [71] and introduces a
and time-series problems in science and engineering have SVM-based prediction model to predict the severity of
been reported by many researchers all around the world tunnel squeezing. However, the difference is that we con-
[2, 3, 30, 31, 40, 41, 52, 76–78, 80, 81, 84–88, sider the effects of the percentage strain (e). There are
91, 92, 94, 96–101]. These methods have been used by several commonly considered predictor variables in this
researchers in the areas of geotechnical [15, 55, 66, 98] and field, which are H, K, D, K and SSR. It seems that there is a
tunnel engineering [75, 81, 83] and also to solve problem need to consider effects of other important parameters on
related to tunnel squeezing [50, 65]. To estimate tunnel tunnel squeezing like the percentage strain (e). The men-
squeezing, ML techniques like artificial neural network tioned parameters were rarely used as input parameter in
(ANN), decision tree (DT), naive Bayes (NB) and support the proposed ML classifier models. Table 2 is the list of
vector machine (SVM) have been used in the literature. As commonly used predictors.
an example, Shafiei et al. [65] used and introduced a SVM Additionally, with the deepening of research, opti-
classifier model, which was trained and tested based on 198 mization algorithms are gradually introduced into machine
samples, in particular having two predictor variables learning methods to optimize hyper-parameters, such as
(buried depth, H, and rock tunneling quality index, Q). The whale optimization algorithm (WOA), gray wolf opti-
accuracy of their proposed model is 84.1%. In another mization (GWO), Harris Hawks optimizer (HHO) and
interesting investigation, Sun et al. [71] constructed a moth-flame optimization (MFO). Therefore, various hybrid
multi-class SVM prediction model based on 117 samples. models have gradually formed such as GWO-SVM
There were four predictor variables (H, Q, diameter, D, and [79, 82], WOA-SVM [94], MFO-SVM, GS-SVM [46],
support stiffness, K,) in the multi-class SVM model, and it HHO-SVM [91], WOA-XGBoost, GWO-XGBoost, BO-
was able to receive an accuracy of 88.1%. Zhang et al. XGBoost [64, 101, 100] and SCA-RF [91]. The above
[86, 87] established a classifier ensemble based on 166 research shows that the hybrid model has better perfor-
cases, which includes five different ML classifiers: ANN, mance than a single machine learning method. Therefore,
SVM, DT, k-nearest neighbor (KNN), and NB. The five the whale optimization algorithm is introduced to improve
variables, i.e., H, D, Q, K and strength stress ratio (SSR), the prediction performance of multi-class SVM. Whale
were selected as input parameters for the classifier optimization algorithm (WOA) has simple structure, few
ensemble, and the final accuracy was obtained as 96%. parameters, strong search ability and easy to implement
Huang et al. [35] proposed a hybrid model of SVM mixed [7].
by back-propagation (BP) for identifying squeezing and Finally, an optimized classifier model (WOA-SVM) is
non-squeezing problem based on a total of 180 data sam- proposed to predict the severity of tunnel squeezing based
ples. In the SVM-BP model, the four indicators including on five parameters, that is, buried depth (H), support
H, K, D and Q were considered as model inputs. The stiffness (K), rock tunneling quality index (Q), diameter
accuracy of the SVM-BP model was obtained as 92.11%. (D), and the percentage strain (e). Firstly, we establish a
In addition, other methods and accuracy comparison results database containing above five surrounding rock indicators
are shown in Table1. In light of above discussion, the based on the existing literature and then preprocessing
performance of the combined classifiers/models is higher these data. Then, the WOA-SVM model was trained and
than the single classifier. However, in most of the cases, the tested of tunnel squeezing. This study copes with not only
combined classifier models are complex with the lowest the development of the WOA-SVM model used for the
level of practicality, when the number of classifiers anticipating of squeezing problems, but also the sensitivity
increases. To solve this problem, this article only uses a analysis of predictor variables. Finally, in order to verify
single classifier SVM. SVM has high generalization per- the advantage of the model proposed, an evaluation and
formance and can solve problems like small samples and comparison on the performance of different classifier
high dimensionality [63]. According to the existing models (WOA-SVM, ANN, SVM, and genetic program-
research, we can also found that support vector machines ming, GP) based on the same database were implemented.
have become popular in engineering. Many researchers The performance and accuracy of the mentioned models
have applied support vector machines to tunnel extrusion will be assessed and discussed to select the best model in
prediction. It can be roughly divided into two applications. predicting tunnel squeezing.
On the one hand, it uses SVM regression to predict the
deformation of the tunnel [39, 72, 90]. On the other hand, it
uses SVM classification to determine whether the tunnel

123
Acta Geotechnica (2022) 17:1343–1366 1345

Table 1 Classification comparison of existing prediction models

Author Classifiers Predictors Number of samples Accuracy Number of
classes

Shafiei et al. [65] SVM H, Q 198 84.1% 2

Sun et al. [71] M-SVM H, Q, D, K 117 88.13% 3
Feng & BNs H, Q, D, K, 166 86.65% 2
Jimenez [21] SSR
Azizi et al. [6] BNs H, Q, D, K, 4(Kerman Water 2
SSR Conveyance Tunnel)
Ghasemi & K-NN H, Q, D 115 95% 2
Gholizadeh [23]
Ghasemi & C5.0 H, Q, D 115 94% 2
Gholizadeh [23]
Chen et al. [13] DT H, D, K, 154 93.5% 3
SSR, GC
Zhang et al. [86, 87] the classifier ensemble (BPNN, SVM, DT, H, Q, D, K, 166 96% 2
KNN, LR, MLR, NB) SSR
Huang et al. [35] SVM-BP H, Q, D, K 178 92.11% 2

Table 2 List of commonly used predictors [9, 13, 21]

Parameters Name Formula

H Buried depth /
Q Rock tunneling quality index Q ¼ ðRQD=Jn Þ  ðJr =Ja Þ  ðJw =SRFÞ
pffiffiffiffiffiffiffiffiffiffiffi
D Diameter D ¼ 4A=p
K Support stiffness K ¼ Kc þ Ksb þ Kb
SSR Strength stress ratio SSR ¼ rcm =cH
e Percentage strain e¼u=D
RQD = Rock quality designation, Jn = joint set number, Jr = joint roughness number, Ja = joint alteration number, Jw = joint water reduction
factor, SRF = stress reduction factor. A = the cross-sectional area of tunnel. Kc = the stiffness of shotcrete linings, Ksb = the stiffness of steel
sets, Kb = the stiffness of rock bolts. rcm = Rock mass uniaxial compressive strength. u = tunnel closure.

2 Predictor selection and database between the support pressure and the rock mass deforma-
description tion response [13]. SSR and e are usually used as grading
indicators such as in the research conducted by Jethwa
According to the published literatures, the research group et al. (1984), Barla [8] and Aydan et al. [4, 5].
collected 114 historical cases of tunnel squeezing from In this study, we adopt the classification standard pro-
various locations like Greece, Bhutan, India, Austria, posed by Hoek and Marinos [33]. Therefore, non-squeez-
China, Nepal and Venezuela [1, 6, 18, 20, 32, 51, ing (NS) (with e \ 1%), minor squeezing (MS) (with
62, 67, 71]. There are six parameters in each case where 1% B e \ 2.5%) and severe-to-extreme squeezing (SES)
five of them (K, H, Q, D and e) were set as input variables (with e C 2.5%) were represented by class 0, class 1 and
to predict tunnel squeezing. Among these six parameters, class 2, respectively. A correlation scatter matrix was
H, Q and D are often appeared in empirical formulas, such performed to know more about the used parameters, as
as H ¼ 350Q0:33 and H ¼ 275N 0:33 B0:1 , which are pro- shown in Fig. 1. The diagonal of the matrix presents
posed by Goel and Singh [27, 68]. The three parameters probability distributions for each squeezing class, the lower
reflect the influence of in situ stress, surrounding rock panels show pairwise scatter plots of three classes of
properties and tunnel size on squeezing. The support squeezing data along the axis and the upper triangle pre-
stiffness is selected as the input parameter. The reason is sents the Pearson’s correlation coefficients. It can be
that the support stiffness plays an important role in con- clearly seen that all indicators have no relatively mean-
trolling the excessive deformation caused by the interaction ingful correlation with each other, and there is no clear

123
1346 Acta Geotechnica (2022) 17:1343–1366

Fig. 1 Correlation scatter matrix of cumulative distributions and statistical evaluations for the squeezing database

separation among NS, MS and SES. The mentioned input ‘‘soft margin’’ was introduced into the SVM model. In this
and output parameters will be used in the next stage for way, the optimization objective functions of SVM can be
classification modeling of tunnel squeezing. expressed in the following [45, 46, 73, 91, 94]:
1 Xm
min kwk2 þC l0=1 ðyi ðwT xi þ bÞ  1Þ ð1Þ
w;b 2
3 Concepts of predictive models i¼1

where l0=1 is 0/1 loss function, which can measure the

3.1 Support vector machine (SVM) deviation degree and can be defined as follows:
(
The SVM has high generalization performance and does 1; if Z\0
l0=1 ðZÞ ¼ ð2Þ
not require prior knowledge of specific models; therefore, it 0; otherwise
is widely used to solve problems in different fields, for
example, finance [47], energy [34], hydrological research With the introduction of slack variables ni and penalty
[58], mechanical engineering [16, 48], civil engineering factors C (the regularization constant), the original opti-
[63, 84] and other fields. Of course, SVM is also widely mization problem can be rewritten as follows:
used for tunnel extrusion prediction [38, 65]. The initial 1 X m
concept of SVM is to input the training data set and output min kwk2 þC ni
w;b 2
i¼1
the separating classification decision function with the ð3Þ
largest geometric interval [12, 16, 34, 47–49, 74]. The s:t:yi ðwT xi Þ  1  ni
SVM has been widely used to solve multivariate classifi- ni  0; i ¼ 1; 2; . . .; m
cation and regression problems [45, 54, 58], although it is a
By introducing the Lagrangian multipliers
binary classification model on nature. The advantage of the
(ai  0; ui  0), the Lagrangian function is constructed to
SVM model lies in the ability to transform nonlinear
solve problems with constraints:
problems into linear problems in high-dimensional feature
spaces with the help of kernel functions [54]. 1 X m X m
Lðw; b; a; n; uÞ ¼ kwk2 þC ni þ ai ð1  ni
In practical problems, it is difficult to find a hyperplane 2 i¼1 i¼1
that can separate different categories of samples when the Xm

training sets are nonlinearly separable in the sample space.  yi ðwT x þ bÞÞ  ui n i ð4Þ
i¼1
To solve this problem, there is a need to allow SVM for
making mistakes on some datasets. Therefore, the sense of

123
Acta Geotechnica (2022) 17:1343–1366 1347

When the partial derivative of the above formula to W, (sequential minimal optimization) algorithm [73]. Finally,
b, ni is zero, the Lagrange dual problem can be described as the classification decision function can be described as:
follows: Xm
f ðxÞ ¼ ai yi UK ðx; xi Þ þ b ð8Þ
Xm
1X m X m
i¼1
max ai  ai aj yi yj xTi xj
a
i¼1
2 i¼1 j¼1 X
m
b ¼ y i  ai yi UK ðxi ; xi Þ ð9Þ
X
m
ð5Þ i¼1
s:t: ai y i ¼ 0
i¼1
0  ai  C; i ¼ 1; 2; . . .; m 3.2 Whale optimization algorithm (WOA)
Optimization problems with inequality constraints need
Inspired by the bubble-net attacking technique which is
to meet the following conditions.
humpback whale’s unique predation method, Mirjalili [53]
ai  0; ui  0 suggested the WOA algorithm for solving and optimizing
yi ðwT x þ bÞ  1 þ n0  0 problems. Therefore, the WOA is widely used in energy,
ð6Þ image processing and machine vision, structural opti-
ai ðyi ðwT x þ bÞ  1 þ ni Þ ¼ 0
mization, management and other fields [53]. Humpback
ni  0; ui ni ¼ 0
whales like to hunt a group of krill or small fish near the
To overcome nonlinear classification and clustering water surface. They are gradually evolved a special hunting
issues, it is essential to choose the appropriate kernel method called foam feeding, that’s because they move
function UK ðx; zÞ as a substitute for inner product to con- slowly. Whale can construct a spiral path with a decreasing
struct and solve the convex quadratic programming issue radius by creating bubbles for enforcing fish schools to
[16, 34, 47]. That means Eq. (5) becomes Eq. (7). In this approach the surface and then catching them [43, 57].
way, the input data can be mapped into a high-dimensional WOA concept can be described as (1) encircling prey, (2)
feature spaces [47], as shown in Fig. 2. bubble-net attacking method and (3) search for prey, which
are discussed in detail as follows (in order to distinguish,
1X m X m Xm
min ai aj yi yj UK ðxi ; xj Þ  ai the bold letters in the following formula represent vectors):
a 2
i¼1 j¼1 i¼1
Encircling prey
X
m
ð7Þ
s:t: ai y i ¼ 0 The exact position of prey cannot be easily identified;
i¼1
therefore, the system considers the solution of the current
0  ai  C; i ¼ 1; 2; . . .; m candidate for the target prey [57]. In the next step, after
Then, we will obtain w and b after calculation of the recognizing the best search agent ðX  ; Y  Þ, there is a need
  T  for the other search agents ðX; YÞ to upgrade their locations
optimal solution a a ¼ a1 ; a2 ; :::; am by the SMO

Feature
Map

Separating
Hyperplane

Complex in two dimensions Simple in high-dimensional feature spaces

Fig. 2 Mapping data from two dimensional to three dimensional

123
1348 Acta Geotechnica (2022) 17:1343–1366

using Eqs. 10 and 11. The nearby solutions around the best The shrinking encompassing mechanism and the spiral
or optimized solution can be according to Eqs. 12 and 13 updating location have an equivalent probability to be
(Fig. 3). selected by the humpback whale in the process of position
  updating. The process of simulation can be demonstrated as
 
R ¼ C  X ðtÞ  XðtÞ  ð10Þ follows:
(
X ðtþ1Þ ¼ XðtÞ  A  R ð11Þ Xðtþ1Þ ¼ XðtÞ  A  R; if p\0:5
Xðtþ1Þ ¼ ð15Þ
A ¼ 2a  r1  a ð12Þ R0  ebl  cosð2plÞXðtÞ ; if p  0:5

C ¼ 2r2 ð13Þ where p is an arbitrary number in the range of [1].

where A and C represent the coefficient vectors, and the (3) Search for prey
value of A is restricted to [1]. Parameter of a can be
In order to update the whale places during the explo-
decreased from 2 to 0 in the search process, and it can be
ration phase, the equation of the model is presented as
calculated by a¼22t=Tmax (t and Tmax represent the cur-
follows:
rent number and the maximum number of iterations,
respectively). Factors r1 and r2 are random vectors in the R ¼ jC  Xrand  XðtÞ j
ð16Þ
range of [1]; XðtÞ and X ðtÞ denote the current whale position Xðtþ1Þ ¼ Xrand  A  R
vector and the best whale solution vector (the possible
location of the prey) in the tth iteration, respectively. where Xrand denotes the whale location vector which is
selected randomly.
Bubble-net attacking method
The humpback whales and their bubble-net attacking 3.3 Multilayer perceptron (MLP)
behavior can be mathematically simulated by designing
two procedures, which are shrinking encircling and spiral In this paper, ANN refers to multilayer perceptron (MLP).
updating. The spiral equation is described in the following Multilayer perceptron (MLP) is promoted from a single-
equation: layer perceptron. The main feature is that it has multiple
neuron layers. Generally, the first layer of MLP is called
Xðtþ1Þ ¼ R0  ebl  cosð2plÞXðtÞ ð14Þ the input layer, the middle layer is the hidden layer and the
last layer is the output layer. MLP does not specify the
where the shape of the logarithmic spiral depends on b
number of hidden layers, so the appropriate number of
which is a constant, l is a random vector which distributed
hidden layers can be selected according to actual process-
uniformly within [-1,1]. The distances between the ith
ing requirements. These hidden layers have different
search agent and the target prey are presented by
numbers of hidden neurons. The neurons in each hidden
R0 ¼ jXðtÞ  XðtÞ j.

(X*, Y, Z) (X, Y, Z)
(X*-X, Y, Z)

X*-X (X*, Y, Z*) (X, Y, Z*)

(X*-X, Y, Z*)

(X*-X, Y) (X*, Y) (X, Y) (X*-X, Y, Z*-Z) (X*, Y, Z*-Z) (X, Y, Z*-Z) (X, Y*,Z)

(X*, Y*,Z*)
Y*-Y

(X, Y*, Z*)

(X*-X, Y*) (X*, Y) (X, Y*) (X*, Y, Z*-Z) (X, Y*, Z*-Z) (X, Y*-Y,Z)
(X*-X, Y*, Z*-Z)

(X, Y, Z*)

(X*-X, Y*-Y) (X*, Y*-Y) (X, Y*-Y)

(X*-X, Y*-Y, Z-Z*) (X*, Y*-Y, Z*-Z) (X, Y*-Y, Z*-Z)

Fig. 3 Different vector positions highlighting the best solutions

123
Acta Geotechnica (2022) 17:1343–1366 1349

Fig. 4 Confusion matrix and performance indicators

advantages are: (1) predictions can explain observations.

(2) The prediction is probabilistic, so that the empirical
confidence interval can be calculated. (3) Versatility.

4 Modeling results and discussion

4.1 Evaluation criteria

The ROC (receiver operating characteristic) curve is very

popular in the performance evaluation phase of ML clas-
sifiers [86, 93, 95]. The ROC curve can be presented in a
form of Cartesian coordinate system, in which FPR (false-
(a)
positive rate) and TPR (true-positive rate) represent as the
horizontal axis and the vertical axis, respectively. The key
indicator of performance evaluation in the ROC curve is
the AUC value that is defined as the area under the ROC
curve. The larger AUC values, the higher the classification
accuracy of the model or the better performance. On the
other hand, accuracy and Cohen’s kappa can be also con-
sidered as performance indicators. The Kappa coefficient
measures the effect of classification by evaluating the
consistency between the prediction results of the model and
the actual classification results. A normal range for results
of kappa is in the range of 0–1. If this range is divided into
five different classes, there are: 1) slight consistency
(b) (0 * 0.20), 2) fair consistency (0.21 * 0.40), 3) moderate
consistency (0.41 * 0.60), 4) substantial consistency
Fig. 5 Model decision boundary before and after optimization:
(a) SVM; (b) WOA-SVM
(0.61 * 0.80) and 5) almost perfect consistency
(0.81 * 1.00). In addition to accuracy and Kappa, preci-
layer have the same activation function, and there is no sion, recall and F1 can also be considered as performance
limit to the number of neurons in each layer in the hidden indicators [86, 92]. The mentioned performance indicators
layer and the output layer. (accuracy, Kappa, precision, recall and F1-score) can be
computed based on the confusion matrix, as shown in
3.4 The Gaussian process (GP) Fig. 4. Based on the confusion matrix, MCC also was
introduced as performance indicators. Matthews correla-
GP means Gaussian process classification. The Gaussian tion coefficient is an index used in machine learning to
process is a general supervised learning method for solving measure the classification performance. This indicator
regression and probability classification problems. The considers true positives, true negatives, false positives and
false negatives. It is generally considered to be a relatively

123
1350 Acta Geotechnica (2022) 17:1343–1366

Fig. 6 The whole analysis process of WOA-SVM classifier model

TP  TN  FP  FN
MCC ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðTP þ FPÞðTP þ FNÞðTN þ FPÞðTN þ FNÞ
ð17Þ

4.2 WOA-SVM model development

and validation

Main steps for constructing WOA-SVM model in pre-

dicting tunnel squeezing are as follows:
Step 1: Data preparation: The database collected from
the existing literature has a total number of 114 cases.
The source of the cited cases and the necessary
information are listed in appendix. A. According to the
Fig. 7 Optimization of WOA-SVM with different population values most commonly used division ratio of 80%/20%, based
on the Pareto principle [64, 99, 102], we randomly divide
balanced indicator, and it can be applied even when the dataset into 80% training set and 20% testing set for
sample content of the two categories differs greatly. MCC model development and model validation, respectively
is essentially a correlation coefficient that describes the [71].
actual classification and the predicted classification. Its Step 2: Initializing parameters of the SVM model. There
value range is [-1,1], 1 indicates a perfect prediction of the are several main parameters in the SVM model, includ-
subject, and a value of 0 indicates that the predicted result ing the penalty parameter of the objective function
is not as good as a random prediction, -1 means that the (‘‘C’’), the kernel function and the coefficient of the
predicted classification is completely inconsistent with the kernel function (‘‘g’’). The hyper-parameters ‘‘C’’ and
actual classification. The calculation formula is as follows: ‘‘g’’ need to be optimized by WOA algorithm. In this
research, the kernel function is determined with the help
of the model decision boundary diagram. The SVM

123
Acta Geotechnica (2022) 17:1343–1366 1351

Table 3 The performance of the SVM model optimized with WOA

Type Swarm Accuracy Rank Kappa Rank MCC Rank Total

Training 50 0.9890 5 0.9820 5 0.9823 5 15

Training 80 0.9890 5 0.8606 4 0.9823 5 14
Training 100 0.9890 5 0.9820 5 0.9823 5 15
Training 150 0.9890 5 0.9820 5 0.9823 5 15
Training 200 0.9890 5 0.9820 5 0.9823 5 15
Testing 50 0.9565 5 0.9288 5 0.9316 4 14
Testing 80 0.9130 4 0.8606 4 0.8633 3 11
Testing 100 0.9565 5 0.9288 5 0.9317 5 15
Testing 150 0.9565 5 0.9288 5 0.9317 5 15
Testing 200 0.9130 4 0.8606 4 0.8633 3 11

Observed class Observed class

Class0 Class1 Class2 UA/% Class0 Class1 Class2 UA/%
Class2 Class1 Class0

Class2 Class1 Class0

6 0 0 100 6 0 0 100
Predicted class
Predicted class

1 4 0 80 1 4 0 80

0 0 12 100 0 1 11 91.67
PA/%

PA/%

85.71 100 100 85.71 80 100

Kappa=0.9288, Accuracy=0.9565, Kappa=0.8606, Accuracy=0.913,

MCC=0.9317 MCC=0.8633
(a) (b)
Observed class Observed class
Class0 Class1 Class2 UA/% Class0 Class1 Class2 UA/%
Class2 Class1 Class0

Class2 Class1 Class0

4 0 2 66.67 2 0 4 33.33
Predicted class

Predicted class

0 0 5 0 0 0 5 0

0 0 12 100 0 1 11 91.67
PA/%

PA/%

100 0 63.16 100 0 55

Kappa=0.4188, Accuracy=0.6957, Kappa=0.1544, Accuracy=0.5652,

MCC=0.5227 MCC=0.2095
(c) (d)
Fig. 8 Confusion matrix different prediction methods: a WOA-SVM; b ANN; c SVM; and d GP

123
1352 Acta Geotechnica (2022) 17:1343–1366

(a) WOA-SVM (b) ANN

(c) GP (d) SVM

Fig. 9 Actual and predicted classification results on test datasets

Table 4 Performance of different classifiers for the non-squeezing like linear, polynomial, radial basis function (RBF), and
problems, minor squeezing problems and high squeezing problem sigmoid. According to the model decision boundary
Precision Recall F1-score Precision Recall F1-score diagram in Fig. 5, it is easy and feasible to detect that the
database in this article is close to linearly separable.
WOA-SVM ANN Therefore, the linear kernel was applied to input
NS 0.86 1.00 0.92 0.86 1.00 0.92 parameter mapping for the SVM model.
MS 1.00 0.80 0.89 0.80 0.80 0.80 Step 3: The relevant parameters of the WOA and their
SES 1.00 1.00 1.00 1.00 0.92 0.96 ranges are the constant b, two random number l 2 ½1; 1
SVM GP and r 2 ½0; 1½0; 1. It is necessary to determine and
NS 1.00 0.67 0.80 1.00 0.33 0.50 design the optimal hyper-parameters (C and g) of SVM
MS 0.00 0.00 0.00 0.00 0.00 0.00 using the WOA. Therefore, a WOA-SVM hybrid model
SES 0.63 1.00 0.77 0.55 0.92 0.69 can optimize the ability of the SVM classifier in
predicting tunnel squeezing through WOA algorithm.
The specific optimization process of the proposed WOA-
model transforms the linearly inseparable problem into SVM is shown in Fig. 6.
linearly separable with the help of the kernel functions

123
Acta Geotechnica (2022) 17:1343–1366 1353

(a)
(a)

(b)
Fig. 11 Taylor graph a test sets, b train sets
(b)
Step 4: Fitness evaluation of WOA-SVM model and
determination of the optimal population size. It is
necessary for developing a reliable WOA-SVM model
with the best performance to fix the optimal population
number. This is because swarm size has a significant
impact on the performance of the WOA model. To
search the optimal population number, five different
swarm sizes (i.e., 50, 80, 100, 150 and 200) were
selected and used in the process of model development.
The fitness curve presented in Fig. 7 shows that the
adaptation value changes with the number of iterations.
When the number of iterations is greater than or equal to
80, the fitness values of the five fitness curves generated
by the WOA-SVM model will tend to be stable. Table 3
presents the results of performance evaluation (accuracy
(c) and Kappa) for the optimization WOA-SVM model
based on the training and testing sets. Based on this
Fig. 10 ROC curves and AUC values for different individual
classifiers: a non-squeezing problems; b minor squeezing problems; table and considering all performance indexes, the
c severe-to-extreme squeezing problems optimal population or swarm size was selected as 150
with accuracy = 0.9565 and Kappa = 0.9288.

123
1354 Acta Geotechnica (2022) 17:1343–1366

Feature
D

0 0.2 0.4 0.6 0.8 1

Importance
(a)

class2

Class
Fig. 12 The violin chart presented for different classifier models class1 K
D
4.3 Analysis and comparison of classification Q
performance class0 H

The optimized classifier model based on the training set 0 0.1 0.2 0.3 0.4
needs to be validated based on testing datasets. The test Importance
datasets were randomly selected from the database pre- (b)
pared, i.e., 20% of the total cases (23 test samples). It is
important to mention that they have not participated in the Fig. 13 Variable contribution analysis: a overall analysis; b analysis
training process of the model. We will analyze and com- of variables for non-squeezing problems, minor squeezing problems,
and high squeezing problem
pare classification performance from different perspectives
such as confusion matrix, performance evaluation indica- predicted ones obtained from different classifiers, the
tors, violin graphs and so on. From the confusion matrix,
resultant classification results are demonstrated in Fig. 9.
we can get the accuracy, Kappa, MCC and other perfor- We can see the 23 samples of the test dataset on the hor-
mance evaluation indicators and use then analyzing and izontal axis and the class of the sample on the vertical axis
comparing the classification performance of different
(class0: non-squeezing; class1: minor squeezing; class2:
models on the basis of these evaluation indicators. To severe-to-extreme squeezing). There is a sample with the
examine the accuracy of the WOA-SVM model, the actual class: class 1 in Fig. 9a, which was misclassified as
methods of GP, ANN and SVM were built for classification
class 0, and this sample was defined as case No. 20.
purpose of the same samples. The results of the verification However, there are more than one sample in Fig. 9 (b, c
are shown in Fig. 8, which represents the confusion matrix and d), which was misclassified. The WOA-SVM model is
of four classification models (WOA-SVM, SVM, ANN and
more accurate and safer in predicting the level of tunnel
GP) for testing datasets. It is not difficult to observe that the squeezing.
WOA-SVM classifier demonstrates better performance
The above analysis aims to evaluate the classification
than the other built models. Compared with the other un-
performance of the model as a whole. However, imbal-
optimized classifier models, the WOA-SVM classification anced dataset may have a great impact on the prediction
model has the highest accuracy (approximately 0.9565). In
results of the model, but it is not enough to detect this
addition, the Kappa values obtained for different classifiers influence based on the accuracy rate alone. Therefore,
from high to low are: 0.929 (WOA-SVM), 0.913 (ANN), precision, recall, F1 and ROC curves were also applied and
0.696 (SVM) and 0.565 (GP). In addition to accuracy and
calculated to assess the prediction performance of WOA-
Kappa mentioned above, the number of cases classified SVM, SVM, ANN and GP models. Table 4 tabulates pre-
correctly can be obtained from the main diagonal of the cision, recall and F1-score of different classification mod-
confusion matrix.
els based on non-squeezing (NS), minor squeezing (MS)
The above analysis has shown that the WOA-SVM and severe-to-extreme squeezing(SES). According to this
model has certain advantages. In order to present the dif- table, the WOA-SVM model was able to receive a better
ference between measured tunnel squeezing results and the

123
Acta Geotechnica (2022) 17:1343–1366 1355

Prediction probabilities NOT class 0 class 0

ε <= 0.79
class 0 0.69 0.74

class 1 0.24 227.50 < H <= 3...

0.01
class 2 0.07 4.60 < D <= 5.80
0.01
0.05 < Q <= 0.32
0.00
K > 480.50
0.00

Feature Value
ε 0.77
H 276.00
D 5.00
Q 0.25
K 940.00
Fig. 14 Probabilistic interpretation of the non-squeezing category

Prediction probabilities NOT class 1 class 1

0.79 < ε <= 2.66
class 0 0.09 0.49

class 1 0.52 H > 562.50

0.04
class 2 0.39 9.80 < K <= 26.19
0.03
Q > 0.32
0.02
4.60 < D <= 5.80
0.00
Feature Value
ε 2.33
H 635.00
K 9.81
Q 4.00
D 5.80
Fig. 15 Probabilistic interpretation of the minor squeezing category

performance and higher level of accuracy. Based on the Figure 10b and c demonstrates AUC values of different
above analysis, for the optimized ML classifier, the clas- classifiers based on class 1 and class 2, respectively. The
sification performance of the optimized SVM model was specific values can be obtained from the figure. In Fig. 10,
significantly improved compared to the base model which the AUC values obtained from the WOA-SVM model
is SVM. based on class 0, class 1 and class 2 are 1, 0.93 and 1,
ROC curves and AUC values of different individual respectively. Obviously, the WOA-SVM model is the
classifiers for different classes are shown in Fig. 10. preferred ML classifier for squeezing degree prediction.
According to Fig. 10a, the AUC values based on the class 0 In order to understand the capability of our proposed
were calculated as 1, 0.99, 0.78, 0.93 and 0.94 for WOA- model better, we have drawn Taylor graph for train and test
SVM, ANN, GP and SVM approaches, respectively. sets separately, as shown in Fig. 11. Taylor chart is often

123
1356 Acta Geotechnica (2022) 17:1343–1366

Prediction probabilities NOT class 2 class 2

2.66 < ε <= 4.87
class 0 0.05 0.35

class 1 0.28 9.80 < K <= 26.19

0.03
class 2 0.67 Q <= 0.02
0.02
5.80 < D <= 8.70
0.01
227.50 < H <= 3...
0.01

Feature Value
ε 3.10
K 22.58
Q 0.01
D 6.80
H 337.00
Fig. 16 Probabilistic interpretation of the high squeezing category

used to evaluate the accuracy of a model. Commonly used Therefore, it is particularly important to evaluate the con-
accuracy indicators are MCC, standard deviation and root- tribution of input parameters to the developed model. The
mean-square error (RMSE). Generally speaking, the scat- Shapley Additive Explanations was used to obtain the
tered points in the Taylor diagram represent the model, the importance of predictive variables to WOA-SVM classifi-
radial line represents the MCC, the horizontal and vertical cation model. The calculation formula is shown as [99]:
axis represents the standard deviation and the dashed line X jSj!ðjN j  jSj  1Þ!
represents the root-mean-square error. The Taylor chart is a Ui ¼ ½qs[i ðxs[i Þ  qs ðxs Þ ð18Þ
S N=ðiÞ
jN j!
change from the previous scatter chart, which can only
show two indicators to express the accuracy of the model. where N represents the set of all features in the data set, S is
Similarly, we still can see that the WOA-SVM model is the the set after index i is removed, the importance of feature i
preferred ML classifier for squeezing degree prediction. to the model output is represented by Ui , xs represents the
The above analysis is based on the test set. Below we vector of input features in set S, and the contribution of
will analyze and compare the performance of the model features is calculated with the corresponding function q.
based on all the sample data in this article. The violin In practical applications, the prediction results based on
chart includes a combined specifications of the box plot the predictor variables with high contribution rates to
and the kernel density plot. The main application of this model are more reliable and accurate. There are five fea-
chart is to present the probability density and distribution tures in this work, and the importance of predictor vari-
of datasets. Figure 12 shows the distribution and proba- ables to the WOA-SVM classification model was
bility density of prediction accuracy for different classifier calculated (Fig. 13). It can be intuitively seen that the
models considering all 114 samples. The prediction accu- percentage strain (e) is the most important parameter in
racy of WOA-SVM model is higher than the other classi- predicting tunnel squeezing, followed by K and H param-
fication models, and the distribution of accuracy is more eters. Due to the imbalance dataset of this article, the
concentrated, which sufficiently illustrates that the hybrid contribution of the parameters to model based on different
model (WOA-SVM) has visible advantages in squeezing types of data (class0, class1 and class2) was assessed
prediction. (Fig. 13b) According to Fig. 13, e is still the most influ-
ential parameter on the model for all classes. However, for
4.4 Sensitivity analysis of predictor variables class 1 and class 2, the parameter K ranks second only to e
in the contribution rate rankings, followed by H. For class
The key to predicting tunnel squeezing is the selection of 0, the parameter H ranks second only to e, followed by K.
appropriate input parameters. The research of Huang et al. In summary, the parameters that have important contribu-
[35] showed that the coupling effect of different parameters tion to the WOA-SVM model are: e, K, H and D.
has different effects on tunnel squeezing prediction.

123
Acta Geotechnica (2022) 17:1343–1366 1357

In order to verify the above conclusions, we randomly and the AUC were calculated. The aim of the sensitivity
selected a sample from three different classes, and the analysis of predictor variables is to evaluate the contribu-
probabilistic interpretation of the sample is given in tion of input parameters to the model. The main results of
Figs. 14–16. That means that Figs. 14, 15 and 16 demon- this study are summarized as follows.
strate that the five parameters (e, K, D, Q and H) have
(1) The WOA algorithm can effectively optimize the
different contributions to the prediction of class0, class1
hyper-parameters of the SVM classifier and improve
and class2. Figure 14 presents the process of the sample
its classification performance. The WOA-SVM clas-
selected was considered class0 by WOA-SVM model
sification model has the highest accuracy (approxi-
according to input parameters. According to the informa-
mately 0.9565) than other un-optimized individual
tion in Fig. 14, it can be easily observed that the proba-
classifiers (SVM, ANN and GP). However, the
bilities of the sample to class0, class1 and class2 are 0.69,
model has a good classification effect, even if the
0.24 and 0.07, respectively. Therefore, the final prediction
data are unbalanced.
result is class0 (light squeezing problem), and parameters
(2) The results of the sensitivity analysis indicate that e,
e, H and K are decisive predictor variables, where e plays a
H and K are the best combination of parameters for
decisive role in the prediction results.
WOA-SVM model, where the percentage strain (e) is
Similarly, Fig. 15 displays the process of the sample
the most influential factor on the WOA-SVM model,
selected was judged to be class1. The sample will be
the parameter K ranks second only to e in the
judged to be class0, class1 and class2 with corresponding
contribution rate rankings, followed by H.
probability of 0.09, 0.52 and 0.39, respectively. Finally,
this sample is considered as class1 (moderate squeezing So far, most of the existing forecasting methods can
problem). The decisive predictor variables are differ- distinguish between squeezing and non-squeezing. This
ent from that presented in Fig. 14, and they are e, H and D. article refers to the multi-class SVM proposed by Sun et al.
Nevertheless, parameters e still has the deepest effect on [71] and introduces the whale optimization algorithm to
the proposed model. Figure 16 demonstrates that the optimize the prediction performance of the multi-class
sample will be regarded as class0, class1 and class2 with SVM. Therefore, the WOA-SVM model has higher pre-
corresponding probability of 0.05, 0.28 and 0.67, respec- diction accuracy than empirical methods, ordinary binary
tively. Obviously, this sample is ultimately considered as SVM and multi-class SVM and can predict the severity of
class2 (high squeezing problem), and the percentage strain tunnel squeezing. However, compared with numerical
(e) is the most important parameter for predicting tunnel simulation, the influencing factors considered by this paper
squeezing, followed by the parameters K and H. In other are obviously limited. In addition, in the actual construc-
words, Figs. 14, 15 and 16 illustrate that the parameters e, tion process, it is difficult to obtain more accurate input
K, H and D have a considerable impact on the WOA-SVM parameter values. According to the research of Zhang et al.
model, while e is the most important input parameter [73,74], the prediction performance of the classifier
among them. ensemble model is higher than that of the individual clas-
sifier. In future, other advanced single classifiers can be
introduced to construct a classifier ensemble. On this basis,
5 Conclusion the introduction of suitable optimization algorithms can
greatly improve the prediction accuracy of the model. In
We proposed an optimized classifier model (WOA-SVM) addition, expanding the existing database can improve the
to estimate the potential of tunnel squeezing according to generalization ability of the integrated model.
114 cases. There were five input parameters (H, K, D, Q
and e) considered in the modeling of all ML models in this
study (WOA-SVM, ANN, SVM and GP). In order to assess Appendix
the performance of different classifier models based on the
same database, accuracy, kappa, precision, recall, F1-score See Table 5.

123
1358 Acta Geotechnica (2022) 17:1343–1366

Table 5 Performance of different classifiers at many problems in minor change

No Tunnel Location Rock type H (m) Q D (m) K (MPa) e (%) Reference Class

1 Bhutan Bhutan AGO (adverse 337.000 0.007 6.800 44.760 2.100 Sripad et al. [70] MS
geological)
2 Chameliya Nepal Dolomite 199.700 0.020 5.400 1217.160 4.580 Basnet [10] SES
hydroelectric
project
headrace tunnel
3 ? 172
3 Chameliya Nepal Dolomite 217.500 0.013 5.400 1217.160 25.540 Basnet [10] SES
hydroelectric
project
Headrace tunnel
3 ? 190
4 Chameliya Nepal Brownish 252.200 0.010 5.400 1523.070 12.500 Basnet [10] SES
hydroelectric
project
Headrace tunnel
3 ? 296
5 Chameliya Nepal Foliated phyllite 246.300 0.010 5.400 1523.070 3.800 Kumar [10] SES
hydroelectric
project
Headrace tunnel
3 ? 314
6 Chameliya Nepal Talcose phyllite 283.900 0.008 5.400 1645.380 36.730 Basnet [10] SES
hydroelectric
project
headrace tunnel
3 ? 404
7 Chameliya Nepal Talcose phyllite 284.500 0.008 5.400 1828.980 30.190 Basnet [10] SES
hydroelectric
project
Headrace tunnel
3 ? 420
8 Chameliya Nepal Talcose phyllite 210.800 0.010 5.400 1575.720 18.300 Kumar [44] SES
hydroelectric
project
Headrace tunnel
3 ? 681
9 Chameliya Nepal Talcose phyllite 237.700 0.010 5.400 1575.720 10.960 Basnet [10] SES
hydroelectric
project
Headrace tunnel
3 ? 733
10 Chameliya Nepal Foliated phyllite 230.000 0.015 5.400 1217.160 9.800 Basnet [10] SES
hydroelectric
project
Headrace tunnel
3 ? 764
11 Chameliya Nepal Foliated phyllite 222.600 0.015 5.400 1217.160 1.200 Basnet [10] MS
hydroelectric
project
Headrace tunnel
3 ? 795
12 Chenani–Nashri India Siltstone, silty 727.000 2.287 6.000 5.880 1.700 Kumar [44] MS
escape tunnel claystone
13 Chenani–Nashri India Siltstone, silty 733.000 2.903 6.000 6.250 1.600 Kumar [44] MS
escape tunnel claystone

123
Acta Geotechnica (2022) 17:1343–1366 1359

Table 5 (continued)
No Tunnel Location Rock type H (m) Q D (m) K (MPa) e (%) Reference Class

14 Chenani–Nashri India Siltstone, silty 736.000 2.426 6.000 7.690 1.300 Kumar [44] MS
escape tunnel claystone
15 Chenani–Nashri India Siltstone, silty 690.000 1.650 6.000 9.380 1.600 Kumar [44] MS
escape tunnel claystone
16 Chenani–Nashri India Siltstone 577.000 1.517 13.000 11.110 1.800 Kumar [44] MS
escape tunnel
17 Chibro-Khodri India Soft and plastic 280.000 0.022 3.000 5.960 4.500 Goel et al. [29] SES
black clays
18 Chibro-Khodri India Crushed red shales 280.000 0.050 3.000 9.800 2.800 Hoek [32], SES
Goel et al [28]
19 Chibro-Khodri India Seamy crushed red 680.000 0.050 9.000 9.900 6.000 Goel et al. [29] SES
20 Chibro-Khodri India Soft and plastic 280.000 0.022 9.000 48.560 2.000 Goel et al. [29] MS
black clays
21 Giri-Bata tunnel India Crushed phyllites 200.000 0.020 4.600 2.980 6.200 Goel [26]; Choudhari SES
[14]; Dube [17]
22 Giri-Bata tunnel India Crushed phyllites 325.000 0.030 4.600 2.980 8.750 Goel [26]; Choudhari SES
[14]; Dube [17]
23 Giri-Bata tunnel India Crushed slates 400.000 0.512 4.600 2.980 0.670 Goel [26]; Choudhari NS
[14]; Dube [17]
24 Giri-Bata tunnel India Crushed phyllites 440.000 0.050 4.600 3.970 10.040 Goel [26]; Choudhari SES
[14]; Dube [17]
25 Giri-Bata tunnel India Crushed phyllites 450.000 0.060 4.600 3.970 10.300 Goel [26] Choudhari SES
[14]; Dube [17]
26 Giri-Bata tunnel India Crushed phyllites 240.000 0.120 4.600 3.970 4.500 Goel [26]; Choudhari SES
[14]; Dube [17]
27 Giri-Bata tunnel India Crushed phyllites 400.000 0.030 4.600 3.980 10.430 Goel [26]; Choudhari SES
[14]; Dube [17]
28 Giri-Bata tunnel India Crushed phyllites 400.000 0.050 4.600 3.980 7.610 Goel [26]; Choudhari SES
[14]; Dube [17]
29 Kaligandaki ‘‘A’’ Nepal Graphitic phyllite 620.000 0.008 8.700 14.670 8.500 NEA [56]; Panthi et al. SES
HRT [62]
30 Kaligandaki ‘‘A’’ Nepal Graphitic phyllite 620.000 0.009 8.700 14.670 7.700 NEA [56]; Panthi et al. SES
HRT [62]
31 Kaligandaki ‘‘A’’ Nepal Graphitic phyllite 620.000 0.009 8.700 14.670 8.200 NEA [56]; SES
HRT Panthi et al. [62]
32 Kaligandaki ‘‘A’’ Nepal Graphitic phyllite 575.000 0.007 8.700 21.170 6.000 NEA [56]; Panthi et al. SES
HRT [62]
33 Kaligandaki ‘‘A’’ Nepal Graphitic phyllite 620.000 0.020 8.700 26.100 4.100 NEA [56]; Panthi et al. SES
HRT [62]
34 Kaligandaki ‘‘A’’ Nepal Siliceous phyllites 620.000 0.016 8.700 26.200 4.400 NEA [56]; SES
HRT Panthi et al. [62]
35 Kaligandaki ‘‘A’’ Nepal Graphitic phyllite 600.000 0.018 8.700 26.200 3.900 NEA [56]; SES
HRT Panthi et al. [62]
36 Kaligandaki ‘‘A’’ Nepal Graphitic phyllite 620.000 0.020 8.700 26.200 4.900 NEA [56]; Panthi et al. SES
HRT [62]
37 Kaligandaki ‘‘A’’ Nepal Graphitic phyllite 580.000 0.023 8.700 26.200 3.700 NEA [56]; SES
HRT Panthi et al. [62]
38 Kaligandaki ‘‘A’’ Nepal Graphitic phyllite 600.000 0.023 8.700 28.480 3.200 NEA [56]; SES
HRT Panthi et al. [62]
39 Kaligandaki ‘‘A’’ Nepal Graphitic phyllite 600.000 0.030 8.700 34.480 2.900 NEA [56]; SES
HRT Panthi et al. [62]
40 Kaligandaki ‘‘A’’ Nepal Graphitic phyllite 550.000 0.029 8.700 39.130 2.300 NEA [56]; Panthi et al. MS
HRT [62]

123
1360 Acta Geotechnica (2022) 17:1343–1366

Table 5 (continued)
No Tunnel Location Rock type H (m) Q D (m) K (MPa) e (%) Reference Class

41 Kaligandaki ‘‘A’’ Nepal Graphitic phyllite 550.000 0.025 8.700 39.870 2.400 NEA [56]; MS
HRT Panthi et al. [62]
42 Kaligandaki ‘‘A’’ Nepal Graphitic phyllite 620.000 0.025 8.700 50.800 2.500 NEA [56]; SES
HRT Panthi et al. [62]
43 Kaligandaki ‘‘A’’ Nepal Graphitic phyllite 580.000 0.025 8.700 74.660 1.700 NEA [56]; MS
HRT Panthi et al. [62]
44 Kaligandaki ‘‘A’’ Nepal Graphitic phyllite 600.000 0.023 8.700 90.710 1.400 NEA [56]; MS
HRT Panthi et al. [62]
45 Khara hydroproject India Clay conglomerate 200.000 0.400 6.000 20.000 0.750 Goel et al. [28]; Singh NS
et al. [69]
46 Khara hydroproject India Clay conglomerate 150.000 0.400 6.000 26.190 0.420 Goel et al. [28]; NS
Singh et al. [69]
47 Khimti 1 Nepal Gneiss and sericite 98.000 0.080 4.000 933.000 0.770 Shrestha [67] NS
hydroproject A1 schists
ch475
48 Khimti 1 Nepal Sheared schists 100.000 0.010 4.200 31.720 3.810 Shrestha [67] SES
hydroproject A1
ch500
49 Khimti 1 Nepal Sheared schists 100.000 0.005 4.200 88.960 2.620 Shrestha[67] SES
hydroproject A1
ch515
50 Khimti 1 Nepal Sheared schists 111.000 0.008 4.300 1936.000 0.750 Shrestha [67] NS
hydroproject A1
ch580
51 Khimti 1 Nepal Gneiss and schists 112.000 0.060 4.000 458.000 0.300 Shrestha [67] NS
hydroproject A1
ch665
52 Khimti 1 Nepal schists 212.000 0.040 4.400 5324.000 0.020 Shrestha [67] NS
hydroproject A2
ch1283
53 Khimti 1 Nepal Banded gneiss and 261.000 0.095 4.000 931.000 0.150 Shrestha [67] NS
hydroproject A2 chlorite schists
ch1357
54 Khimti 1 Nepal Gneiss 95.000 0.065 4.000 933.000 0.290 Shrestha [67] NS
hydroproject A2
ch1730
55 Khimti 1 Nepal Gneiss 126.000 0.300 4.000 461.000 0.030 Shrestha [67] NS
hydroproject A2
ch441
56 Khimti 1 Nepal Sericite schists 138.000 0.013 4.000 1934.000 0.190 Shrestha [67] NS
hydroproject A2
ch601
57 Khimti 1 Nepal Gneiss and 198.000 0.140 4.000 934.000 0.290 Shrestha [67] NS
hydroproject A2 chlorite schists
ch895
58 Khimti 1 Nepal Gneiss and schists 130.000 0.200 5.000 936.000 0.340 Shrestha [67] NS
hydroproject A3
ch15
59 Khimti 1 Nepal Gneiss and schists 276.000 0.250 5.000 940.000 0.770 Shrestha [67] NS
hydroproject A3
ch200
60 Khimti 1 Nepal Gneiss and schists 276.000 0.280 5.000 652.000 0.560 Shrestha [67] NS
hydroproject A3
ch210

123
Acta Geotechnica (2022) 17:1343–1366 1361

Table 5 (continued)
No Tunnel Location Rock type H (m) Q D (m) K (MPa) e (%) Reference Class

61 Khimti 1 Nepal Schists 140.000 0.009 4.000 430.000 0.800 Shrestha [67] NS
hydroproject A3
ch220
62 Khimti 1 Nepal Gneiss and schist 284.000 0.090 5.000 68.550 1.240 Shrestha [67], MS
hydroproject A3 Panthi [60]
ch235
63 Khimti 1 Nepal Gneiss and sericite 300.000 0.090 5.000 664.290 0.280 Shrestha [67] NS
hydroproject A3 schists
ch340
64 Khimti 1 Nepal schists 300.000 0.050 5.000 1430.000 0.180 Shrestha [67] NS
hydroproject A3
ch345
65 Khimti 1 Nepal Gneiss and schists 158.000 0.230 4.100 650.000 0.320 Shrestha [67] NS
hydroproject A3
ch59
66 Khimti 1 Nepal Sericite schists 112.000 0.006 4.000 71.280 1.190 Shrestha [67] MS
hydroproject A4
ch1013
67 Khimti 1 Nepal Clay-filled sheared 112.000 0.008 4.000 651.000 0.100 Shrestha [67] NS
hydroproject A4
ch1045
68 Khimti 1 Nepal Gneiss and sericite 225.000 0.140 4.000 1430.000 0.240 Shrestha [67] NS
hydroproject A4 schists
ch503
69 Khimti 1 Nepal Chlorite sericite 218.000 0.070 4.000 739.000 0.140 Shrestha [67] NS
hydroproject A4 gneiss
ch550
70 Khimti 1 Nepal Banded gneiss 114.000 0.470 4.000 648.000 0.030 Shrestha [67] NS
hydroproject A4
ch852
71 Khimti 1 Nepal Banded gneiss 114.000 0.600 4.000 556.000 0.240 Shrestha [67] NS
hydroproject A4
ch876
72 Khimti 1 Nepal Gneiss 112.000 0.008 4.000 936.000 0.200 Shrestha [67] NS
hydroproject A4
ch974
73 Loktak hydro India — 300.000 0.023 4.600 7.710 7.000 Goel et al. [28] SES
74 Maneri stage I India — 350.000 0.500 5.800 2.530 7.900 Goel et al. [28] SES
75 Maneri stage I tunnel India Sheared 450.000 0.310 5.800 5.100 4.830 Goel [26]; Jethwa [36] SES
metabasics
76 Maneri stage I tunnel India Crushed quartzite 750.000 0.500 5.800 8.100 4.140 Goel [26]; Jethwa [36] SES
77 Maneri stage I tunnel India Sheared 700.000 0.300 5.800 9.810 4.830 Goel [26]; Jethwa [36] SES
metabasics
78 Maneri stage I tunnel India Siliceous phyllites 550.000 1.700 5.800 9.810 2.660 Goel [26]; Jethwa [36] SES
79 Maneri stage I tunnel India Foliated 635.000 4.000 5.800 9.810 2.330 Goel [26]; Jethwa [14] MS
metabasics
80 Maneri stage I tunnel India Siliceous phyllites 650.000 4.120 5.800 9.810 2.070 Goel [26]; Jethwa [14] MS
81 Maneri stage II India Sheared 285.000 0.100 7.000 9.790 2.870 Goel [26]; Choudhari SES
tunnel metabasics [14]
82 Maneri stage II India Sheared 410.000 0.300 7.000 9.790 2.800 Goel [26]; SES
tunnel metabasics Choudhari [14]
83 Maneri stage II India Metavolcanic 415.000 0.880 7.000 9.790 2.190 Goel [26]; Choudhari MS
tunnel [14]
84 Maneri stage II India Metavolcanic 500.000 1.000 7.000 9.790 2.640 Goel [26]; SES
tunnel Choudhari [14]

123
1362 Acta Geotechnica (2022) 17:1343–1366

Table 5 (continued)
No Tunnel Location Rock type H (m) Q D (m) K (MPa) e (%) Reference Class

85 Maneri stage II India Metavolcanic 480.000 0.800 2.500 9.840 2.880 Goel [26]; Choudhari SES
tunnel [14]
86 Maneri stage II India Metavolcanic 510.000 0.880 2.500 9.840 2.420 Goel [26]; Choudhari MS
tunnel [14]
87 Maneri-Bhali India Fractured quartzite 225.000 3.600 4.800 1000.000 0.060 Singh et al. [69]; Goel NS
hydroproject et al. [29]
88 Maneri-Bhali stage I India Fractured quartzite 350.000 0.500 4.800 25.320 7.900 Hoek [32], SES
Goel et al. [29]
89 Maneri-Uttarkashi India Laminated 800.000 2.500 4.800 48.990 8.900 Hoek [32], SES
power metabasics Goel et al. [29]
90 Maneri-Uttarkashi India Sheared 340.000 1.800 4.800 500.000 0.400 Goel et al. [29] NS
power metabasics
91 Maneri-Uttarkashi India Foliated 550.000 5.100 4.800 1600.000 0.050 Goel et al. [29] NS
power metabasics
92 Nathpa Jhakri-HRT India Quartz mica 700.000 0.417 11.000 7.430 3.500 Kumar [44] SES
schist, schistose
quartzites and
amphibolites
93 Nathpa Jhakri-HRT India Quartz mica 600.000 0.250 11.000 9.140 3.500 Kumar [44] SES
schist, schistose
quartzites and
amphibolites
94 Nathpa Jhakri-HRT India Quartz mica 700.000 0.333 11.000 9.140 3.500 Kumar [44] SES
schist, schistose
quartzites and
amphibolites
95 Nathpa Jhakri-HRT India Quartz mica 750.000 0.333 11.000 9.140 3.500 Kumar [44] SES
schist, schistose
quartzites, and
amphibolites
96 Nathpa Jhakri-HRT India Quartz mica 300.000 0.001 11.000 16.500 6.000 Kumar [44] SES
schist, schistose
quartzites and
amphibolites
97 Nathpa Jhakri-HRT India Quartz mica 400.000 0.003 11.000 17.000 6.000 Kumar [44] SES
schist, schistose
quartzites and
amphibolites
98 Nathpa Jhakri-HRT India Quartz mica 800.000 0.194 11.000 17.140 3.500 Kumar [44] SES
schist, schistose
quartzites and
amphibolites
99 Nathpa Jhakri-HRT India Quartz mica 850.000 0.056 11.000 20.400 5.000 Kumar [44] SES
schist, schistose
quartzites and
amphibolites
100 Nathpa Jhakri-HRT India Quartz mica 600.000 0.033 11.000 33.330 3.000 Kumar [44] SES
schist, schistose
quartzites and
amphibolites
101 Noonidih colliery India Weak coal 450.000 0.590 7.000 9.670 3.000 Jethwa [36] SES
Tala hydro-HRT

123
Acta Geotechnica (2022) 17:1343–1366 1363

Table 5 (continued)
No Tunnel Location Rock type H (m) Q D (m) K (MPa) e (%) Reference Class

102 Tala HRT, Bhutan Bhutan amphibolites and 337.000 0.080 6.800 14.090 2.200 Sripad et al. [70] MS
quartzites
in thin bands
103 Tala HRT, Bhutan Bhutan occurrences): 337.000 0.011 6.800 16.050 3.800 Sripad et al. [70] SES
104 Tala HRT, Bhutan Bhutan completely 337.000 0.006 6.800 22.580 3.100 Sripad et al. [70] SES
sheared,
105 Tala HRT, Bhutan Bhutan highly weathered 337.000 0.006 6.800 36.360 2.200 Sripad et al. [70] MS
biotite schist
associated with
banded gneiss,
106 Tehri dam project India Argillaceous 220.000 0.800 12.000 32.890 0.380 Goel et al. [28]; Singh NS
phyllite et al. [69]
107 Udhampur rail tunnel India Claystone, silty 300.000 0.033 6.500 10.000 3.000 Kumar [44] SES
(T1) claystone
108 Udhampur rail tunnel India Claystone, silty 280.000 0.031 6.500 11.540 2.600 Kumar [44] SES
(T1) claystone
109 Udhampur rail tunnel India Claystone, silty 280.000 0.042 6.500 12.500 2.400 Kumar [44] MS
(T1) claystone
110 Udhampur rail tunnel India Claystone, silty 285.000 0.063 6.500 12.800 2.500 Kumar [44] SES
(T1) claystone
111 Udhampur rail tunnel India Claystone, silty 270.000 0.125 6.500 15.910 2.200 Kumar [44] MS
(T1) claystone
112 Udhampur rail tunnel India Claystone, silty 280.000 0.083 6.500 29.330 1.500 Kumar [44] MS
(T1) claystone
113 Udhampur rail tunnel India Claystone, silty 312.000 0.094 6.500 34.670 1.500 Kumar [44] MS
(T1) claystone
114 Upper Krishna India Banded schists 52.000 15.000 13.000 16.670 0.180 Goel et al. [28]; NS
project Singh et al. [69]

Acknowledgements This research was funded by the National Sci- 6. Azizi F, Koopialipoor M, Khoshrou H (2019) Estimation of rock
ence Foundation of China (42177164) and the Innovation-Driven mass squeezing potential in tunnel route (case study: Kerman
Project of Central South University (No. 2020CX040). water conveyance tunnel). Geotech Geol Eng 37(3):1671–1685.
https://doi.org/10.1007/s10706-018-0714-5
7. Bansal S, Rattan M (2019) Design of cognitive radio system and
comparison of modified whale optimization algorithm with
References whale optimization algorithm. Int J Inf Technol. https://doi.org/
10.1007/s41870-019-00346-2
1. Ajalloeian R, Moghaddam B, Azimian A (2017) Prediction of 8. Barla G (2001) Tunnelling under squeezing rock conditions.
rock mass squeezing of T4 tunnel in Iran. Geotech Geol Eng Mechanics—Advances in Geotechnical Engineering and Tun-
35(2):747–763. https://doi.org/10.1007/s10706-016-0139-y nelling, 169–268. http://scholar.google.com/scholar?hl=
2. Armaghani DJ, Harandizadeh H, Momeni E, Maizir H, Zhou J en&btnG=Search&q=intitle:
(2021a) An optimized system of GMDH-ANFIS predictive Tunnelling?under?squeezing?rock?conditions#0
model by ICA for estimating pile bearing capacity. Artif Intell 9. Barton N, Lien R, Lunde J (1974) Engineering classification of
Rev, 1–38 rock masses for the design of tunnel support. Rock Mech
3. Armaghani DJ, Yagiz S, Mohamad ET, Zhou J (2021) Predic- Felsmechanik Mécanique Des Roches 6(4):189–236. https://doi.
tion of TBM performance in fresh through weathered granite org/10.1007/BF01239496
using empirical and statistical approaches. Tunnell Undergr 10. Basnet CB (2013) Evaluation on the squeezing phenomenon at
Space Technol 118:104183 the headrace tunnel of Chameliya Hydroelectric Project, Nepal
4. Aydan O, Akagi T, Kawamoto T (1993) The squeezing potential 11. Bhasin R, Grimstad E (1996) The use of stress-strength rela-
of rocks around tunnels; theory and prediction. Rock Mech Rock tionships in the assessment of tunnel stability. Tunnell Undergr
Eng 26(2):137–163. https://doi.org/10.1007/BF01023620 Space Technol 11(1):93–98
5. Aydan Ö, Akagi T, Kawamoto T (1996) The squeezing potential 12. Chapelle O, Haffner P, Vapnik VN (1999) Support vector
of rock around tunnels: theory and prediction with examples machines for histogram-based image classification. IEEE Trans
taken from Japan. Rock Mech Rock Eng 29(3):125–143. https:// Neural Netw 10(5):1055–1064
doi.org/10.1007/BF01032650

123
1364 Acta Geotechnica (2022) 17:1343–1366

13. Chen Y, Li T, Zeng P, Ma J, Patelli E, Edwards B (2020) splines approach. Bull Eng Geol Env 77(2):489–500. https://doi.
Dynamic and probabilistic multi-class prediction of tunnel org/10.1007/s10064-016-0937-8
squeezing intensity. Rock Mech Rock Eng 53(8):3521–3542. 32. Hoek E (2001) Big tunnels in bad rock 2000 Terzaghi Lecture.
https://doi.org/10.1007/s00603-020-02138-8 ASCE J Geotech Geoenviron Eng 127(9):726–740
14. Choudhari JB (2007) Closure of underground opening in jointed 33. Hoek E, Marinos P (2000) Predicting tunnel squeezing problems
rocks. PhD Thesis, IIT Roorkee, Roorkee, India in weak heterogeneous rock masses. Tunnels and Tunnelling
15. Dai Y, Khandelwal M, Qiu Y, Zhou J, Monjezi M, Yang P International, 1–20. http://www.rockscience.com/hoek/refer
(2022) A hybrid metaheuristic approach using random forest and ences/H2000d.pdf
particle swarm optimization to study and evaluate backbreak in 34. Hu G, Xu Z, Wang G, Zeng B, Liu Y, Lei Y (2021) Forecasting
open-pit blasting. Neural Comput Appl. https://doi.org/10.1007/ energy consumption of long-distance oil products pipeline based
s00521-021-06776-z on improved fruit fly optimization algorithm and support vector
16. Du M, Zhao Y, Liu C, Zhu Z (2021) Lifecycle cost forecast of regression. Energy. https://doi.org/10.1016/j.energy.2021.
110 kV power transformers based on support vector regression 120153
and gray wolf optimization. Alex Eng J 60:5393–5399. https:// 35. Huang Z, Liao M, Zhang H, Zhang J, Ma S (2020) Predicting
doi.org/10.1016/j.aej.2021.04.019 the tunnel surrounding rock extrusion deformation based on
17. Dube AK (1979) Geomechanical evaluation of tunnel stability SVM-BP model with incomplete data. Mod Tunnel Technol
under failing rock conditions in a Himalayan Tunnel. Depart- (S1), https://doi.org/10.13807/j.cnki.mtt.2020.S1.017.
ment of Civil Engineering, University of Roorkee, Roorkee, 36. Jethwa JL (1981) Evaluation of rock pressures in tunnels
India through squeezing ground in lower Himalayas, University of
18. Dwivedi RD, Goel RK, Singh M, Viladkar MN, Singh PK Roorkee, Roorkee, India
(2019) Prediction of ground behaviour for rock tunnelling. Rock 37. Jimenez R, Recio D (2011) A linear classifier for probabilistic
Mech Rock Eng 52(4):1165–1177. https://doi.org/10.1007/ prediction of squeezing conditions in Himalayan tunnels. Eng
s00603-018-1673-0 Geol 121:101–109. https://doi.org/10.1016/j.enggeo.2011.05.
19. Dwivedi RD, Singh M, Viladkar MN, Goel RK (2013) Predic- 006
tion of tunnel deformation in squeezing grounds. Eng Geol 38. Kang Y, Wang J (2010a) A support-vector-machine-based
161:55–64. https://doi.org/10.1016/j.enggeo.2013.04.005 method for predicting large-deformation in rock mass. 7th
20. Farhadian H, Nikvar-Hassani A (2020) Development of a new International Conference on Fuzzy Systems and Knowledge
empirical method for Tunnel Squeezing Classification (TSC). Discovery, FSKD 2010, 1176–1180. https://doi.org/10.1109/
Q J Eng GeolHydrogeol. https://doi.org/10.1144/qjegh2019-108 FSKD.2010.5569148
21. Feng X, Jimenez R (2015) Predicting tunnel squeezing with 39. Kang, Y., & Wang, J. (2010b). A support-vector-machine-based
incomplete data using Bayesian networks. Eng Geol method for predicting large-deformation in rock mass. Pro-
195:214–224. https://doi.org/10.1016/j.enggeo.2015.06.017 ceedings - 2010 7th International Conference on Fuzzy Systems
22. Frough O, Torabi SR, Yagiz S (2015) Application of RMR for and Knowledge Discovery, FSKD 2010, 1176–1180. https://doi.
estimating rock-mass–related TBM utilization and performance org/10.1109/FSKD.2010.5569148
parameters: a case study. Rock Mech Rock Eng 40. Khandelwal M (2011) Blast-induced ground vibration prediction
48(3):1305–1312 using support vector machine. Eng Comput 27(3):193–200
23. Ghasemi E, Gholizadeh H (2019) Prediction of squeezing 41. Khandelwal M, Monjezi M (2013) Prediction of backbreak in
potential in tunneling projects using data mining-based tech- open-pit blasting operations using the machine learning method.
niques. Geotech Geol Eng 37(3):1523–1532. https://doi.org/10. Rock Mech Rock Eng 46(2):389–396
1007/s10706-018-0705-6 42. Kimura F, Okabayashi N, Kawamoto T (1987) Tunnelling
24. Ghiasi V, Ghiasi S, Prasad A (2012) Evaluation of tunnels under through squeezing rock in two large fault zones of the enasan
squeezing rock condition. J Eng Des Technol 10(2):168–179. tunnel II. Rock Mech Rock Eng, 151–166
https://doi.org/10.1108/17260531211241167 43. Kotary DK, Nanda SJ, Gupta R (2021) A many-objective whale
25. Gioda G, Cividini A (1996) Numerical methods for the analysis optimization algorithm to perform robust distributed clustering
of tunnel performance in squeezing rocks. Rock Mech Rock Eng in wireless sensor network. Appl Soft Comput 110:107650.
29(4):171–193. https://doi.org/10.1007/BF01042531 https://doi.org/10.1016/j.asoc.2021.107650
26. Goel R (1994) Correlations for predicting support pressures and 44. Kumar N (2002) Rock mass characterization and evaluation of
closures in tunnels. Ph.D. thesis, Nagpur University, Nagpur, supports for tunnels in Himalaya. PhD Thesis, IIT Roorkee,
India Roorkee, India
27. Goel RK, Jethwa JL, Paithankar AG (1995) Tunnelling through 45. Li E, Yang F, Ren M, Zhang X, Zhou J, Khandelwal M (2021)
the young Himalayas—a case history of the Maneri-Uttarkashi Prediction of blasting mean fragment size using support vector
power tunnel. Eng Geol 39(1–2):31–44. https://doi.org/10.1016/ regression combined with five optimization algorithms. J Rock
0013-7952(94)00002-J Mech Geotech Eng. https://doi.org/10.1016/j.jrmge.2021.07.013
28. Goel RK, Jethwa JL, Paithankar AG (1995a) Tunnelling through 46. Li E, Zhou J, Shi X, Armaghani DJ, Yu Z, Chen X, Huang P
the young Himalayas—a case history of the Maneri-Uttarkashi (2021) Developing a hybrid model of salp swarm algorithm-
power tunnel. Eng Geol 39(1–2):31–44 based support vector machine to predict the strength of fiber-
29. Goel RK, Jethwa JL, Paithankar AG (1995b) Indian experiences reinforced cemented paste backfill. Eng Comput
with Q and RMR systems. Tunn Undergr Space Technol 37(4):3519–3540
10(1):97–109 47. Liu M, Luo K, Zhang J, Chen S (2021) A stock selection
30. Goh ATC, Zhang W (2012) Reliability assessment of stability of algorithm hybridizing grey wolf optimizer and support vector
underground rock caverns. Int J Rock Mech Min Sci regression. Expert Syst Appl. https://doi.org/10.1016/j.eswa.
55:157–163. https://doi.org/10.1016/j.ijrmms.2012.07.012 2021.115078
31. Goh ATC, Zhang W, Zhang Y, Xiao Y, Xiang Y (2018) 48. Liu Y, Wang L, Gu K (2021) A support vector regression
Determination of earth pressure balance tunnel-related maxi- (SVR)-based method for dynamic load identification using
mum surface settlement: a multivariate adaptive regression heterogeneous responses under interval uncertainties. Appl Soft
Comput. https://doi.org/10.1016/j.asoc.2021.107599

123
Acta Geotechnica (2022) 17:1343–1366 1365

49. Lyu F, Fan X, Ding F, Chen Z (2021) Prediction of the axial 66. Shi XZ, Zhou J, Wu BB, Huang D, Wei W (2012) Support
compressive strength of circular concrete-filled steel tube col- vector machines approach to mean particle size of rock frag-
umns using sine cosine algorithm-support vector regression. mentation due to bench blasting prediction. Trans Nonferr
Compos Struct. https://doi.org/10.1016/j.compstruct.2021. Metals Soc China Eng Ed 22(2):432–441. https://doi.org/10.
114282 1016/S1003-6326(11)61195-3
50. Mahdevari S, Torabi SR (2012) Prediction of tunnel conver- 67. Shrestha GL (2005) Stress induced problems in Himalayan
gence using Artificial Neural Networks. Tunn Undergr Space tunnels with special reference to squeezing. In: Faculty of
Technol 28(1):218–228. https://doi.org/10.1016/j.tust.2011.11. Engineering Science and Technology Department of Geology
002 and Mineral Resources Engineering: Vol. Doctoral t (Issue
51. Majumder D, Viladkar MN, Singh M (2017) A multiple-graph November). https://ntnuopen.ntnu.no/ntnu-xmlui/handle/11250/
technique for preliminary assessment of ground conditions for 248703
tunneling. Int J Rock Mech Min Sci 100:278–286. https://doi. 68. Singh B, Jethwa JL, Dube AK, Singh B (1992) Correlation
org/10.1016/j.ijrmms.2017.10.010 between observed support pressure and rock mass quality.
52. Mehrdanesh A, Monjezi M, Khandelwal M, Bayat P (2021) Tunnell Undergr Space Technol Incorporat Trenchless
Application of various robust techniques to study and evaluate 7(1):59–74. https://doi.org/10.1016/0886-7798(92)90114-W
the role of effective parameters on rock fragmentation. Eng 69. Singh M, Singh B, Choudhari J (2007) Critical strain and
Comput, 1–11 squeezing of rock mass in tunnels. Tunn Undergr Space Technol
53. Mirjalili S, Mirjalili SM, Saremi S, Mirjalili S (2020) Whale 22(3):343–350. https://doi.org/10.1016/j.tust.2006.06.005
optimization algorithm: Theory, literature review, and applica- 70. Sripad SK, Raju GD, Singh Rajbal, Khazanchi RN (2007)
tion in designing photonic crystal filters. Stud Comput Intell. Instrumentation of underground excavations at Tala hydroelec-
https://doi.org/10.1007/978-3-030-12127-3_13 tric project in Bhutan. In: Singh R, Sthapak AK (eds) Pro-
54. Mohammadi B, Mehdizadeh S (2020) Modeling daily reference ceedings international workshop on experiences and
evapotranspiration via a novel approach based on support vector construction of Tala hydroelectric project Bhutan, 14–15 June,
regression coupled with whale optimization algorithm. Agric New Delhi, India, pp 269–282
Water Manag. https://doi.org/10.1016/j.agwat.2020.106145 71. Sun Y, Feng X, Yang L (2018) Predicting tunnel squeezing
55. Monjezi MKM (2013) Prediction of backbreak in open-pit using multiclass support vector machines. Adv Civil Eng.
blasting operations using the Machine Learning Method. https://doi.org/10.1155/2018/4543984
389–396. https://doi.org/10.1007/s00603-012-0269-3 72. Tian Z, Qiao C, Teng W, Liu K (2004) Method of predicting
56. NEA (2002) Geology and geotechnical report, volume IV-A tunnel deformation based on support vector machines. China
andgeological drawings and exhibits, volume V-C, in project- Railway Sci (01)
completion report, N. E. Authority, Kaligandaki ‘‘A’’ Hydro- 73. Vapnik V (1995) The nature of statistical learning theory.
electric Project, Syanga, Nepal Springer, Berlin
57. Okwu MO, Tartibu LK (2021) Whale Optimization Algorithm 74. Vapnik V, Izmailov R (2021) Reinforced SVM method and
(WOA). Stud Comput Intell 927:53–60. https://doi.org/10.1007/ memorization mechanisms. Pattern Recogn 119:108018. https://
978-3-030-61111-8_6 doi.org/10.1016/j.patcog.2021.108018
58. Pai P-F, Hong W-C (2007) A recurrent support vector regression 75. Wang SM, Zhou J, Li CQ, Armaghani DJ, Li XB, Mitri HS
model in rainfall forecasting. Hydrol Process 21(6):819–827. (2021) Rockburst prediction in hard rock mines developing
https://doi.org/10.1002/hyp bagging and boosting tree-based ensemble techniques. J Cent
59. Panet M (1996) Two case histories of tunnels through squeezing South Univ 28(2):527–542
rocks. Rock Mech Rock Eng 29(3):155–164. https://doi.org/10. 76. Xu H, Zhou J, Asteris GP, Jahed Armaghani D, Tahir MM
1007/BF01032652 (2019) Supervised machine learning techniques to the prediction
60. Panthi KK (2011) Effectiveness of post-injection grouting in of tunnel boring machine penetration rate. Appl Sci 9(18):3715
controlling leakage: a case study. J Water, Energy Environ. 77. Yang HQ, Li Z, Jie TQ, Zhang ZQ (2018) Effects of joints on
8:14–18 the cutting behavior of disc cutter running on the jointed rock
61. Panthi KK (2014) Predicting tunnel squeezing: a discussion mass. Tunn Undergr Space Technol 81:112–120. https://doi.org/
based on two tunnel projects. 2013. https://doi.org/10.3126/hn. 10.1016/j.tust.2018.07.023
v12i0.9027 78. Yang HQ, Xing SG, Wang Q, Li Z (2018) Model test on the
62. Panthi KKÃ, Nilsen B (2007) Uncertainty analysis of tunnel entrainment phenomenon and energy conversion mechanism of
squeezing for two tunnel cases from Nepal Himalaya. Int J Rock flow-like landslides. Eng Geol 239:119–125. https://doi.org/10.
Mech Mining Sci 44:67–76. https://doi.org/10.1016/j.ijrmms. 1016/j.enggeo.2018.03.023
2006.04.013 79. Yang HQ, Zeng YY, Lan YF, Zhou XP (2014) Analysis of the
63. Parsa P, Naderpour H (2021) Shear strength estimation of excavation damaged zone around a tunnel accounting for
reinforced concrete walls using support vector regression geostress and unloading. Int J Rock Mech Min Sci 69:59–66.
improved by Teaching–learning-based optimization, Particle https://doi.org/10.1016/j.ijrmms.2014.03.003
Swarm optimization, and Harris Hawks Optimization algo- 80. Yang H, Wang Z, Song K (2020) A new hybrid grey wolf
rithms. J Build Eng. https://doi.org/10.1016/j.jobe.2021.102593 optimizer-feature weighted-multiple kernel-support vector
64. Qiu Y, Zhou J, Khandelwal M, Yang H, Yang P, Li C (2021) regression technique to predict TBM performance. Eng Comput.
Performance evaluation of hybrid WOA - XGBoost, GWO - https://doi.org/10.1007/s00366-020-01217-2
XGBoost and BO - XGBoost models to predict blast - induced 81. Yang J, Liu Y, Yagiz S, Laouafa F (2021) An intelligent pro-
ground vibration. Eng Comput. https://doi.org/10.1007/s00366- cedure for updating deformation prediction of braced excavation
021-01393-9 in clay using gated recurrent unit neural networks. J Rock Mech
65. Shafiei A, Parsaei H, Dusseault MB (2012)Rock squeezing Geotech Eng. https://doi.org/10.1016/j.jrmge.2021.07.011
prediction by a support vector machine classifier. 46th US Rock 82. Yang J, Yagiz S, Liu YJ, Laouafa F (2021) a comprehensive
Mechanics / Geomechanics Symposium 2012, 489–503. https:// evaluation of machine learning algorithms on application to
doi.org/10.13140/RG.2.1.3836.3040 predict TBM performance. Undergr Space. https://doi.org/10.
1016/j.undsp.2021.04.003l

123
1366 Acta Geotechnica (2022) 17:1343–1366

83. Yang H, Wang Z, Song K (2020) A new hybrid grey wolf 94. Zhou J, Li E, Yang S, Wang M, Shi X, Yao S, Mitri HS (2019)
optimizer - feature weighted—multiple kernel—support vector Slope stability prediction for circular mode failure using gradi-
regression technique to predict TBM performance. Eng Comput. ent boosting machine approach based on an updated database of
https://doi.org/10.1007/s00366-020-01217-2 case histories. Saf Sci 118(2018):505–518. https://doi.org/10.
84. Yagiz S, Karahan H (2011) Prediction of hard rock TBM pen- 1016/j.ssci.2019.05.046
etration rate using particle swarm optimization. Int J Rock Mech 95. Zhou J, Qiu Y, Zhu S, Armaghani DJ, Li C, Nguyen H, Yagiz S
Min Sci 48(3):427–433 (2021) Optimization of support vector machine through the use
85. Zhang H, Shi Y, Yang X, Zhou R (2021) A firefly algorithm of metaheuristic algorithms in forecasting TBM advance rate.
modified support vector machine for the credit risk assessment Eng Appl Artif Intell 97:104015. https://doi.org/10.1016/j.
of supply chain finance. Res Int Bus Financ. https://doi.org/10. engappai.2020.104015
1016/j.ribaf.2021.101482 96. Zhou J, Li EM, Wang MZ, Chen X, Shi XZ, Jiang LS (2019b)
86. Zhang J, Huang Y, Ma G, Yuan Y, Nener B (2021) Automating Feasibility of stochastic gradient boosting approach for evalu-
the mixture design of lightweight foamed concrete using multi- ating seismic liquefaction potential based on SPT and CPT case
objective firefly algorithm and support vector regression. histories. J Performance Constr Facil 33(3)
Cement Concr Compos. https://doi.org/10.1016/j.cemconcomp. 97. Zhou J, Li XB, Mitri HS (2015) Comparative performance of six
2021.104103 supervised learning methods for the development of models of
87. Zhang J, Li D, Wang Y (2020) Predicting tunnel squeezing hard rock pillar stability prediction. Nat Hazards 79(1):291–316
using a hybrid classifier ensemble with incomplete data. Bull 98. Zhou J, Li XB, Mitri HS (2016) Classification of rockburst in
Eng Geol Env 79:3245–3256. https://doi.org/10.1007/s10064- underground projects: comparison of ten supervised learning
020-01747-5 methods. J Comput Civil Eng, 30(5)
88. Zhang W, Zhang R, Wu C, Goh ATC, Lacasse S, Liu Z, Liu H 99. Zhou J, Chen C, Wang M, Khandelwal M (2021) Proposing a
(2020) State-of-the-art review of soft computing applications in novel comprehensive evaluation model for the coal burst lia-
underground excavations. Geosci Front 11(4):1095–1106 bility in underground coal mines considering uncertainty factors.
89. Zhang W, Li H, Li Y, Liu H, Chen Y, Ding X (2021b) Appli- Int J Min Sci Technol 31(5):799–812
cation of deep learning algorithms in geotechnical engineering: 100. Zhou J, Qiu Y, Khandelwal M, Zhu S, Zhang X (2021)
a short critical review. Artif Intell Rev, 1–41 Developing a hybrid model of Jaya algorithm-based extreme
90. Zhang W, Wu C, Zhong H, Li Y, Wang L (2021) Prediction of gradient boosting machine to estimate blast-induced ground
undrained shear strength using extreme gradient boosting and vibrations. Int J Rock Mech Min Sci 145:104856
random forest based on Bayesian optimization. Geosci Front 101. Zhou J, Qiu Y, Zhu S, Armaghani DJ, Khandelwal M, Moha-
12(1):469–477 madd ET (2021) Estimation of the TBM advance rate under hard
91. Zhao H (2005) Predicting the surrounding deformations of rock conditions using XGBoost and Bayesian optimization.
tunnel using support vector machine. Chin J Rock Mech Eng Underground Space 6(5):506–515. https://doi.org/10.1016/j.
24(4): 649–652. https://doi.org/10.3321/j.issn:1000-6915.2005. undsp.2020.05.008
04.017 102. Zhou J, Qiu Y, Armaghani DJ, Zhang W, Li C, Zhu S, Tarinejad
92. Zhou J, Dai Y, Khandelwal M, Monjezi M, Yu Z (2021) Per- R (2021d) Predicting TBM penetration rate in hard rock con-
formance of hybrid SCA-RF and HHO-RF models for predicting dition: a comparative study among six XGB-based metaheuristic
backbreak in open-pit mine blasting operations. Nat Resour Res techniques. Geosci Front 12(3):101091. https://doi.org/10.1016/
30(6):4753–4771. https://doi.org/10.1007/s11053-021-09929-y j.gsf.2020.09.020
93. Zhou J, Huang S, Wang M, Qiu Y (2021b) Performance eval-
uation of hybrid GA-SVM and GWO-SVM models to predict Publisher’s Note Springer Nature remains neutral with regard to
earthquake-induced liquefaction potential of soil: a multi-dataset jurisdictional claims in published maps and institutional affiliations.
investigation. Eng Comput

Authors and Affiliations

Jian Zhou1 • Shuangli Zhu1 • Yingui Qiu1 • Danial Jahed Armaghani2 • Annan Zhou3 • Weixun Yong1

& Jian Zhou

[email protected]; [email protected] 1
School of Resources and Safety Engineering, Central South
& Yingui Qiu University, Changsha 410083, China
[email protected] 2
Department of Urban Planning, Engineering Networks and
Shuangli Zhu Systems, Institute of Architecture and Construction, South
[email protected] Ural State University, 76, Lenin Prospect,
Danial Jahed Armaghani Chelyabinsk 454080, Russia
[email protected] 3
Civil and Infrastructure Engineering Discipline, School of
Annan Zhou Engineering, Royal Melbourne Institute of Technology
[email protected] (RMIT), Melbourne, VIC 3001, Australia

Weixun Yong
[email protected]

123