Engineering with Computers (2022) 38 (Suppl 3):S1835–S1843

https://doi.org/10.1007/s00366-021-01358-y

ORIGINAL ARTICLE

Stability evaluation of dump slope using artificial neural network

and multiple regression
Ashutosh Kumar Bharati1 · Arunava Ray1 · Manoj Khandelwal2 · Rajesh Rai1 · Ashok Jaiswal1

Received: 21 December 2020 / Accepted: 16 February 2021 / Published online: 9 March 2021
© The Author(s), under exclusive licence to Springer-Verlag London Ltd. part of Springer Nature 2021

Abstract
The present paper focuses on designing an artificial neural network (ANN) model and a multiple regression analysis (MRA)
that could be used to predict factor of safety of dragline dump slope. To implement these two models, the dataset was utilized
from the numerical simulation results of dragline dump slopes, wherein 216 dragline dump slope models were simulated
using a numerical modeling technique employed with the finite element method. The finite element model was incorporated
a combination of three geometrical parameters, namely, coal-rib height (Crh), dragline dump slope height (Sh), and dragline
dump slope angle (Sa) of the dump slope. The predicted results derived from the MRA and ANN models were compared
with the results obtained from the numerical simulation of the dump slope models. Moreover, to compare the validity of
both the models, various performance indicators, such as variance account for (VAF), determination coefficient (R2), root
mean square error (RMSE), and residual error were calculated. Based on these performance indicators, the ANN model has
shown a higher prediction accuracy than the MRA model. The study reveals that the ANN model developed in this research
could be handy in designing the dragline dump slopes at the preliminary stage.

Keywords Multiple regression analysis · Artificial neural network · Finite element method · Dragline dumps

1 Introduction possible and the height increases up to its maximum work-

ing height. Therefore, possesses instability threats to the
Dump slope is an integral part of the opencast mines, which working environment. Apart from the two mentioned fac-
comprises the fragmented overburden rock material. The tors, geotechnical and geometrical factors also significantly
most economical overburden dumping method is inter- contribute to the stability of dump slopes. Any improper
nal dumping, wherein the fragmented overburden rock is combination of these affecting factors may lead to the fail-
dumped inside the pit [1]. In the case of mechanized open- ure of the dragline dump slope. There are some cases of
cast mines, dragline proves to be very efficient and economi- dump failure in the past, which accounted for a huge loss
cal in side-casting the immediate overburden material. The in terms of men, machinery and materials to the mines as
working of dragline mining is shown in Fig. 1. As the work- studied by Kainthola [2], Sharma and Roy [3], and Dash [4].
ing depth increases, the height along with the steepness of Therefore, dump slope stability is very important because
dragline dump increases, because the overburden material it may endanger the safety of men and machinery as well as
is dragged by the dragline from the bench on which the dra- imposes a temporary yet prolonged halt in the production of
gline is sitting; thus, it can make the bench slope as steep as the opencast mine [5–7].
Dump slope stability analysis is of utmost importance,
as it assists in replicating a design for the most favorable
* Manoj Khandelwal dragline dump slope profile. There are various methods to
[email protected];
[email protected] analyze the stability of the dump slopes, and most of them
utilize the factor of safety (FOS) as an indicator to estimate
1
Department of Mining Engineering, Indian Institute the slope stability [8]. The limit equilibrium method (LEM)
of Technology (BHU) Varanasi, Varanasi 221005, UP, India is a traditional and most widely used method to calculate
2
School of Engineering, Information Technology and Physical FOS of the slopes [9–13]. However, they are coupled with
Sciences, Federation University Australia, Ballarat, certain presumptions and limitations, which lead to an
VIC 3350, Australia

13
Vol.:(0123456789)

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

S1836 Engineering with Computers (2022) 38 (Suppl 3):S1835–S1843

Fig. 1  Schematic diagram of the dragline mining method

imprecise outcome. Nowadays, numerical modeling has variables and the calculated FOS as an output value. The pre-
appeared as a revolutionary method in solving complicated dicted outcomes of ANN and MRA have been compared with
and practical slope stability projects effectively, in which the the calculated results from numerical simulation to identify
finite element method (FEM) has contributed as an effective the accuracy and utility of both the developed models.
approach to calculate accurate results of the slope stabil-
ity problems [14, 15]. The FEM primarily helps in design-
ing complex structures, such as dragline dump profiles and 2 Stability analysis of dragline dump slope
employs the geotechnical parameters, such as unit weight,
cohesion, internal friction angle, etc. Afterward, boundary 2.1 Dragline dump slope
conditions and loading conditions are applied to simulate
the problems by imposing the real condition on the model. Dragline mining comes under the mechanized opencast
Many researchers have used the FEM approach to analyze mining method, where removal and handling of immedi-
the dump slope stability [1, 2, 16]. Therefore, in geotechni- ate overburden rock are performed by dragline machines,
cal engineering, FEM has been accepted as a more accurate which forms a unique dump slope profile as shown in Fig. 2.
method in solving complex practical problems. After removal of overburden rock, coal is excavated from
In recent years, various machine learning tools have been beneath, and a coal-rib barrier is left on the dump side to
implemented successfully in several slope stability projects, provide support to the dragline dump slope, which helps in
whether it was for natural or artificial slopes. The artificial holding an additional volume of overburden material [29].
neural network (ANN) being one of them, has been consid-
erably used in solving various slope stability cases [17–23].
Also, there are several documented cases of the use of mul-
tiple regression analysis (MRA) for stability analysis [19, 24,
25]. The multiple regression analysis (MRA) is a statistical
technique that mainly derives a relationship between the out-
put variable and the input variables. This technique has also
been found to be useful for solving slope stability problems.
The present study aims to utilize various machine learning
tools to solve stability problems related to the overburden
dumps. In this regard, ANN and MRA models have been
developed to predict FOS of the homogeneous dragline dump
slope. The ANN and MRA models have been designed using
the Statistica10 software, which has been used by previous
researchers [26–28]. These models have used 216 simu-
lated results data of dragline dump slopes using the FEM
based software, in which three parameters are taken as input Fig. 2  Schematic diagram of dragline dump slope

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Engineering with Computers (2022) 38 (Suppl 3):S1835–S1843 S1837

The geometrical parameters used in this study have been its input variable into a target variable [24, 28, 33]. The
shown in Fig. 2. ANN model comprises hundreds of single units, neurons,
and respective weights combinedly formed neural structures.
Since it processes the information, that’s why it is also called
2.2 Calculation of FOS
processing elements (PE). The PE is an equation which
forms a relationship between independent and dependent
The dataset of 216 dragline dump models, which have been
variables. There are numerous types of neural networks that
previously analyzed by Bharati [29], is used in this study.
have been designed so far and are still being designed, but
A combination of three geometrical parameters was used to
all the models can be delineated by their neuron’s transfer
design the model, such as coal-rib height (Crh), dragline dump
function, the training algorithm, and the connection formula
slope height (Sh), and dragline dump slope angle (Sa). Here,
[34].
Crh ranges from 10 to 20 m at an interval of 2 m, Sh ranges
from 25 to 45 m at an interval of 4 m, and Sa ranges from
30° to 50° at an interval of 5°. The FEM based numerical
simulation technique is used to analyze the models, because
3.1 Development of artificial neural network
FEM has shown an effective approach in solving slope stability
models
problems, such as it is accurate in conveniently designing the
The simulation data of 216 numerical models of dragline
complex and irregular geometries, various types of material
dump slope were utilized, which were analyzed using finite
properties can be included in a model, it can model progressive
element method based RS2 V9.0 a Rocscience software,
failures and calculates the deformation at a different location
with different geometrical parameters by Bharati [29]. The
at the same time, and it does not requires any pre-assumptions
simulation results are used to construct artificial neural net-
about the location and shape of critical slip surfaces. In the
work architecture. ANN model is designed to predict the
study, FEM has implemented the shear strength reduction
factor of safety of the homogeneous dragline dump slope by
(SSR) technique to calculate FOS of the dump models. In the
using its geometrical parameters as input variables. Three
SSR technique, simulation is executed in a series of decreasing
geometrical parameters affecting the FOS of dragline dump
or increasing Ftrial, in which actual strength properties, such
slopes are included in the ANN model; namely, the coal-rib
as internal friction angle and cohesion are altered as per the
height (Crh), dragline dump slope height (Sh), and dragline
mathematical formula mentioned in Eqs. (1) and (2).
dump slope angle (Sa) as input variables and the FOS is
[ ]
1 opted as an output parameter.
𝜑trial = tan−1 trial tan𝜑 , (1)
F The Statistica10 software has been used to develop the
ANN model. This software consists of two types of neural
1
ctrial = c, (2)
F trial
where, φtrial is the altered internal friction angle and ctrial is
the altered cohesion on each Ftrial value, whereas φ and c
are the actual internal friction angle and cohesion values,
respectively. The Ftrial is considered as FOS at which the
dump slope model has just failed.

3 Artificial neural network

Artificial neural network is a modern computing technique

used to simulate data in a manner the human brain analyzes
and processes any information. It comprises an intercon-
nected assembly of artificial neurons that pass on the infor-
mation through the tendons, which exist in the neuron. ANN
is a powerful tool for the simulation of data used primarily
when the relationship between the independent and target
variables is unknown, and it can easily identify the existing
correlation between both the variables [30–32]. The neural
network architecture puts in the picture how it transforms Fig. 3  Schematic diagram of the developed neural network

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

S1838 Engineering with Computers (2022) 38 (Suppl 3):S1835–S1843

network architecture viz. Multilayer perceptron (MLP) and used as a validation data set to construct the ANN model [28].
Radial basis function (RBF) [35]. In this study, MLP is used, Since, several studies have observed that the division of data
as it is the most popular feedforward neural network archi- has a considerable impact on the ANN model’s outcome [37],
tecture [22]. The neural network architecture comprises two and the ANNs are also incapable of providing accurate results
important tools, training algorithm and activation function if the data lies outside the range of training data. Therefore,
of neurons. The Broyden-Fletcher-Goldfarb-Shanno (BFGS) the training and validation data set must characterize the same
training algorithm is used, which requires a smaller number population. The statistical parameters of the applied training
of iterations due to its fast convergence rate and smart search and validation data sets for ANN are shown in Table 1.
criteria [26, 35]. The activation function of the hidden and In designing the neural network architecture, the selection
output layer neurons is very crucial in deciding the desired of the number of hidden layers, and the number of neurons
neural network architecture, and it transforms the transmitted within the hidden layer is highly decisive in the performance
information from the previous layer to the subsequent layers of ANN models. It has been observed that a single hidden
by applying its mathematical function. To devise the best layer can approximate most of the functions in the network,
ANN model, several networks can be trained simultaneously that is why a single hidden layer was kept in the ANN model
by opting a range of the number of neurons within the hid- [38]. In this software, a range of 3–11 neurons has been
den layers and the desired transfer function to the hidden and opted to train the ANN model. There are two more layers,
output layer of the model. Each neural network includes three i.e. input and output layers other than the hidden layer. The
layers such as; the input layer, one or more hidden layers, number of neurons in these layers is constrained to the num-
and an output layer. To assess the performance of the neural ber of input and output variables, respectively [39]. A sche-
network, a sum of square (SOS) error function is used, which matic diagram of the developed neural network architecture
is the sum of squares of differences between the actual and with three layers is depicted in Fig. 3. The transfer function
predicted outputs of the training data set. used for the hidden and output layer is sigmoidal, which is
The dragline dump simulation results data were divided generally applicable in most of neural networks [19, 40, 41].
into two parts, such as training and validation data sets, to
build the ANN model and to determine the performance
validity of the ANN model, respectively [36]. In the present
research, out of the 216 datasets, 78% (169 samples) were
used as a training data set, and rest 22% (47 samples) were

Table 1  Statistical data of input Input and output variables Data sets Statistical parameters Range
and output variables of the
artificial neural network model Mean Std. D Min Max

Crh (Coal rib height) Training 15.101 3.415 10 20 10–20

Validation 14.638 4.301 10 20
Sh (Dragline dump slope height) Training 34.799 6.830 25 45 25–45
Validation 35.723 7.539 25 45
Sa (Dragline dump slope angle) Training 40.272 6.897 30 50 30–50
Validation 39.021 6.861 30 50
FOS (Factor of safety) Training 1.182 0.135 0.91 1.53 0.91–1.53
Validation 1.202 0.157 0.89 1.48 0.89–1.48

Table 2  Summary of trained MLP neural networks

Network name Training Validation Training error Validation error Training Error function Hidden activa- Output
perfor- performance algorithm tion function activation
mance function

MLP 3-10-1 0.999585 0.999215 0.000008 0.000018 BFGS SOS Logistic Logistic
MLP 3-6-1 0.999467 0.999161 0.000010 0.000018 BFGS SOS Logistic Logistic
MLP 3-8-1 0.999569 0.999243 0.000008 0.000017 BFGS SOS Logistic Logistic
MLP 3-10-1 0.999607 0.999173 0.000007 0.000019 BFGS SOS Logistic Logistic
MLP 3-6-1 0.999248 0.999162 0.000014 0.000018 BFGS SOS Logistic Logistic

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Engineering with Computers (2022) 38 (Suppl 3):S1835–S1843 S1839

Training Validation

1.46 1.46

1.36 1.36
Measured FOS

Measured FOS
1.26 1.26

1.16 1.16

1.06 1.06

0.96 0.96
R² = 0.9996 R² = 0.9991
0.86 0.86
0.88 1.08 1.28 1.48 0.88 1.08 1.28 1.48
Predicted FOS Predicted FOS
(a) (b)
All

1.46

1.36
Measured FOS

1.26

1.16

1.06

0.96
R² = 0.999
0.86
0.88 0.98 1.08 1.18 1.28 1.38 1.48
Predicted FOS
(c)

Fig. 4  Comparison of predicted and measured FOS values for a Training, b Validation, and c All datasets

4 Multiple regression analysis compared with the calculated or measured output values by
using different performance indicators discussed in Sect. 5.3.
Generally, the objective of regression analysis is to project In this study, MRA is performed using Crh (coal-rib
the character of interconnections among the various input height), Sh (dragline dump slope height), and Sa (dragline
and output variables. In addition to this, multiple regression dump slope angle) as independent variables and FOS (factor
analysis (MRA) is applicable for more than one input vari- of safety) as the output variable. The simulation results of
able. It also provides the most appropriate equation show- 216 models are used to perform the MRA. The MRA models
ing the relation between independent and criterion variables shows the following relation:
[42, 43]. The equation is further used to predict the output FOS = 2.37129 − 0.017579 ∗ Crh − 0.008179 ∗ Sh − 0.015881 ∗ Sa.
for particular values of the input variables and its validity is
(3)

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

S1840 Engineering with Computers (2022) 38 (Suppl 3):S1835–S1843

Table 3  Connection weights of the ANN model activation function for the hidden and output layer has
Hidden Connection weights proved to be the best ANN model for the prediction of FOS
neurons of dragline dump slope. A comparison between the calcu-
Crh Sh Sa FOS/output
lated FOS values from the numerical simulation and pre-
1 − 0.002 − 0.331 0.581 8.613 dicted values from the ANN model for training, validation,
2 0.302 0.459 0.578 − 16.354 and all dataset are shown in Fig. 4a–c, respectively. From the
3 − 2.991 1.946 2.694 − 0.939 figures, it can be easily said that the calculated and predicted
4 − 0.251 0.272 − 0.644 7.401 FOS values are very close to each other, and the R2 value is
5 − 6.750 − 1.124 − 1.198 12.434 also very close to unity. Thus, it can be confirmed that the
6 − 0.287 − 0.628 − 0.707 − 11.248 FOS of the homogeneous dragline dump slope could be eas-
7 − 0.206 0.702 0.749 − 3.545 ily predicted using the trained ANN model with admissible
8 4.249 3.143 14.570 − 16.824 precision at the beginning stage of designing the dragline
9 2.356 2.742 6.272 − 22.288 dump slope profile.
10 1.030 0.102 2.985 − 1.371 The weights of output and input hidden layer connec-
tion for the different combinations of the three geometrical
parameters of dragline dump slope models are shown in
Table 4  Relative importance of input variables Table 3. By considering the connection weights, the output-
Input variables Importance (%)
hidden connection weight is divided into the factors attached
with every input or independent variables, which assists in
Coal-rib height (Crh) 25 estimating the relative importance of all the input variables
Dragline dump height (Sh) 26 [44].
Dragline dump slope angle (Sa) 49 It is essential to identify the significance of all the three
Total 100 input parameters on the result or output of the ANN model
[45, 46]. Therefore, to perceive the impact of input variables,
the relative importance is calculated, which is helpful in
5 Result and discussion guiding for the selection of optimum values of the input vari-
ables that form dragline dump holding a maximum volume
5.1 Artificial neural network models of overburden material with safety. The relative importance
of the three geometrical parameters is evaluated by using the
Based on the multiple training results, the best five results Garson equation based on the neural net weight matrix. The
have been obtained, as listed in Table 2. The ANN model Garson equation [44] is depicted in Eq. (4), and the absolute
with MLP 3-10-1 is found with the least error (0.000007) value of the connection weights is used in the equation.
and with the maximum coefficient of determination
(R 2 = 99.96%) value. Therefore, the ANN model com-
prises one hidden layer with ten neurons, and the sigmoidal

Table 5  Summary of the MRA for the 216 dragline dump models
Summary of multiple regression analysis

Regression statistics
Multiple R 0.988402933
R2 0.976940358
Adjusted R2 0.976614042
Standard error 0.021090906
Observations 216
Geometrical Standardized regression Standard error of b* Raw regression Standard error of b t-stat P value
parameters coefficient (b*) coefficient (b)

Intercept 2.371290 0.012901 183.8032 0.00

Crh − 0.436382 0.010429 − 0.017579 0.000420 − 41.8416 0.00
Sh − 0.406042 0.010429 − 0.008179 0.000210 − 38.9325 0.00
Sa − 0.788442 0.010429 − 0.015881 0.000210 − 75.5982 0.00

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Engineering with Computers (2022) 38 (Suppl 3):S1835–S1843 S1841

All where i is the input variables, j is the neurons within the hid-
den layer, Qi is the relative importance of the ith input vari-
able on the output or dependent variable, ­OWj,i is the weight
1.46
of output variable for the jth neuron, I­ Wj,i is the weight of
1.36 jth neuron from the ith input variable of hidden layer and n
denotes the number of neurons. The relative importance of
Measured FOS

1.26 the three input variables on the output of the ANN model is
computed and listed in Table 4. From the table, it is clearly
1.16 evident that Sa has the maximum influence with a relative
importance of 49% on the stability of the dragline dump
1.06
slope. Therefore, Sa is the most influential parameter among
0.96 the three.
R² = 0.9769
0.86 5.2 Multiple regression analysis
0.88 0.98 1.08 1.18 1.28 1.38 1.48
Predicted FOS The MRA model has three input parameters as Crh, Sh, and
Sa and an output or criterion parameter FOS. The deter-
Fig. 5  Comparison of predicted and measured FOS values of MRA mination coefficient of the model is shown in Table 5. The
MRA model is showing a close proximity between the pre-
Table 6  Performance indicators of the developed ANN model dicted and measured FOS values in Fig. 5, as the value of
R2 VAF RMSE
R2 (0.9769) is close to unity. The summary of the statistical
results of MRA has been given in Table 5, showing stand-
ANN ardized regression coefficient (b*) and ‘raw’ or ‘unstandard-
Training 0.9996 99.92144 0.003784 ized’ regression coefficient (b). The b* depicts about the
Validation 0.9991 99.83432 0.006112 contribution or influence of independent variables (Crh, Sh
MRA 0.9769 97.69405 0.020894 and Sa) on the dependent variable (FOS), whereas b is used
as a coefficient of independent variables for the calculation
of output value [35]. Geometrical parameter Sa has the high-
∑n ��� � ∑3 �
�
� � �
�
est b* coefficient value and the Sh has the least b* coeffi-
�IWj,i �∕ i=1 �IWj,i � �OWji �
j=1 � � � � � � cient. Therefore, Sa is the maximum influential parameter
Qi = ∑ �∑ �� � �� , (4)
3 n �
� IW
� ∑3 �
� ∕ �IW
� �
� � OW
�
� to the FOS of the dump slope. By considering the statisti-
i=1 j=1 � j,i � i=1 � j,i � � j,i �
cal regression outcome, it can be confirmed that the MRA
model can be utilized in designing the preliminary stage of
dragline dump profiles.

ANN MRA
(a) 0.20 (b) 0.20

0.15 0.15
Residual error of FOS
Residual error of FOS

0.10 0.10

0.05 0.05

0.00 0.00

-0.05 -0.05

-0.10
-0.10
-0.15
-0.15
-0.20
-0.20 0.00 50.00 100.00 150.00 200.00
0.00 50.00 100.00 150.00 200.00
Sample no. Sample no.

Fig. 6  Variation of residual error values predicted by a ANN and b MRA models

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

S1842 Engineering with Computers (2022) 38 (Suppl 3):S1835–S1843

5.3 Validation of performances of the MRA and ANN model was acquired from the numerical simulation results
models of 216 dragline dump models, wherein the three geometrical
parameters were used, namely; coal-rib height, slope height,
To compare the prediction performance of the MRA and and slope angle of the dragline dump. By incorporating the
ANN models, the determination coefficient (­ R2) between the three geometrical parameters, an ANN and an MRA model
predicted and the measured values could be used as a dis- were developed to assess and evaluate the stability of the
tinguishable indicator. Several other performance parameters dragline dump slope. Based on the observations, some cru-
like variance accounted for (VAF) and root mean square error cial conclusions have been drawn are as follows:
(RMSE) is also calculated to observe the performance of the
prediction accuracy of the predictive models devised in this • The performance of ANN and MRA models were com-
work, as implemented by Erzin [47–49] and both the param- pared with FOS values acquired from the numerically
eters have been delineated by Eqs. (5) and (6), respectively. simulated results. These models can be used for the pre-
The suggested models are considered better if the value of diction of FOS as a function of Crh, Sh and Sa and has
VAF is close to 100% and RMSE is close to 0. conferred satisfactory results with coefficient of determi-
[ ( )] nation value 0.9996 and 0.9769, respectively.
var y − ̂y • To examine the prediction accuracy of MRA and ANN
VAF = 1 − × 100, (5)
var(y) models, several performance indicators such as RMSE,
VAF and variation of residual error were used. The
√ RMSE for both the models were 0.020894 and 0.003784
1 ∑N ( )2 and the VAF values were 97.69 and 99.92, respectively.
RMSE = yi − ̂
yi , (6)
N i=1 • The analysis of residual error shows the variation of the
predicted values from the simulation results for the MRA
where, y and ŷ are the calculated or measured and the pre-
and ANN models, which ranges from − 0.53 to + 0.54
dicted values, respectively. Whereas var signifies as variance
and − 0.020 to + 0.015, respectively, thus confirms ANN
and N denotes the number of model samples.
has lower variation of residual error.
The calculated values of VAF and RMSE for the MRA
• Based on ANN and MRA analysis it was observed that
and ANN models are mentioned in Table 6, which shows
among the three most sensitive geometrical parameters
better performance of the ANN model in comparison to
Sa is the most crucial parameter, while Crh and Sh are
the MRA models, because the VAF and RMSE values of
almost equally important but less crucial in comparison
ANN are comparatively higher and lower, respectively,
to Sa.
with respect to MRA. The neural network model performs
• Based on these performance indicators, it has been
better because of its higher degree of robustness and fault
observed that the ANN model has a higher prediction
tolerance due to the presence of numerous processing neu-
accuracy than MRA. The level of accuracy attained by
rons and each with a greater level of interconnections [42].
both the machine learning tools have proved that artificial
An additional method to compare the MRA and ANN
neural network architecture is a handy tool in estimating
model is a residual error, which primarily shows the devia-
the safe dragline dump slope profile.
tion of predicted values from the measured or calculated
values of FOS, wherein the distance between the predicted
and measured values are computed. The variation of residual
References
error for the MRA and ANN model is shown in Fig. 6. After
that, a comparison of the MRA and ANN model has been 1. Rai R, Kalita S, Gupta T, Shrivastva BK (2012) Sensitivity analy-
conducted, which depicts the deviation interval (− 0.020 sis of internal dragline dump stability: finite element analysis.
to + 0.015) of the ANN model is relatively smaller than the Geotech Geol Eng 30(6):1397–1404
deviation interval (− 0.053 to + 0.054) of the MRA model. 2. Kainthola A, Verma D, Gupte SS, Singh TN (2011) A coal mine
dump stability analysis—a case study. Geomaterials 1(01):1
Thus, it also confirms that the ANN model performs com- 3. Sharma S, Roy I (2015) Slope failure of waste rock dump at
paratively better than the MRA model. Jayant opencast mine, India: a case study. Int J Appl Eng Res
10(13):33006–33012
4. Dash AK (2019) Analysis of accidents due to slope failure in
Indian opencast coal mines. Curr Sci 117(2):304
6 Conclusion 5. Speck RC, Huang SL, Kroeger EB (1993) Large-scale slope move-
ments and their affect on spoilpile stability in Interior Alaska. Int
In this study, efforts were made to develop an ANN and J Surf Min Reclam 7(4):161–166
MRA model that can be applied for the estimation of FOS 6. Kasmer O, Ulusay R, Gokceoglu C (2006) Spoil pile instabilities
with reference to a strip coal mine in Turkey: mechanisms and
of the dragline dump slopes. The data used to devise the assessment of deformations. Environ Geol 49(4):570–585

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Engineering with Computers (2022) 38 (Suppl 3):S1835–S1843 S1843

7. Singh Y, Sharma V, Pandita SK, Bhat GM, Thakur KK, Kotwal 29. Bharati AK, Ray A, Rai R, Shrivastava BK (2020) Safety chart for
SS (2014) Investigation of landslide at Sangaldan near tunnel-47, the identification of stability of internal dragline dumps. J Inst Eng
on Katra-Qazigund railway track, Jammu and Kashmir. J Geol Soc (India): Series D 101(2):173–186. https​://doi.org/10.1007/s4003​
India 84(6):686–692 3-020-00225​-2
8. Manouchehrian A, Gholamnejad J, Sharifzadeh M (2014) Devel- 30. Khandelwal M, Singh TN (2011) Predicting elastic proper-
opment of a model for analysis of slope stability for circular mode ties of schistose rocks from unconfined strength using intelligent
failure using genetic algorithm. Environ Earth Sci 71(3):1267–1277 approach. Arab J Geosci 4:435–442. https​://doi.org/10.1007/s1251​
9. Fellenius W (1936) Calculation of stability of earth dam. In: Trans- 7-009-0093-6
actions. 2nd Congress Large Dams, Washington, vol. 4, pp 445–462 31. Agatonovic-Kustrin S, Beresford R (2000) Basic concepts of artifi-
10. Bishop AW (1955) The use of the slip circle in the stability analysis cial neural network (ANN) modeling and its application in pharma-
of slopes. Geotechnique 5(1):7–17 ceutical research. J Pharm Biomed Anal 22(5):717–727
11. Morgenstern NU, Price VE (1965) The analysis of the stability of 32. Monjezi M, Ahmadi Z, Varjani AY, Khandelwal M (2013) Back-
general slip surfaces. Geotechnique 15(1):79–93 break prediction in the Chadormalu iron mine using artificial
12. Spencer E (1967) A method of analysis of the stability of neural network. Neural Comput Appl 23:1101–1107. https​://doi.
embankments assuming parallel inter-slice forces. Geotechnique org/10.1007/s0052​1-012-1038-7
17(1):11–26 33. Gupta M, Jin L, Homma N (2004) Static and dynamic neural net-
13. Janbu N (1973) Slope stability computations. Publication of: Wiley works: from fundamentals to advanced theory. Wiley
(John) and Sons, Incorporated 34. Khandelwal M, Marto A, Fatemi SA et al (2018) Implementing an
14. Griffiths DV, Lane PA (1999) Slope stability analysis by finite ele- ANN model optimized by genetic algorithm for estimating cohe-
ments. Geotechnique 49(3):387–403 sion of limestone samples. Eng Comput 34:307–317. https​://doi.
15. Cai F, Ugai K (2003) Reinforcing mechanism of anchors in slopes: org/10.1007/s0036​6-017-0541-y
a numerical comparison of results of LEM and FEM. Int J Numer 35. StatSoft I (2011) Statistica (data analysis software system), 10
Anal Meth Geomech 27(7):549–564 36. Twomey JM, Smith AE (1997). In: Kartam N, Flood I, Garrett JH
16. Gupta T, Rai R, Jaiswal A, Shrivastva BK (2014) Sensitivity analysis (eds) Artificial neural networks for civil engineers: fundamentals
of coal rib stability for internal mine dump in opencast mine by finite and applications. ASCE, New York, pp 44–64
element modelling. Geotech Geol Eng 32(3):705–712 37. Tokar SA, Johnson PA (1999) Rainfall-runoff modelling using arti-
17. Alimohammadlou Y, Najafi A, Gokceoglu C (2014) Estimation of ficial neural networks. J Hydrologic Eng 4(3):232–239
rainfall-induced landslides using ANN and fuzzy clustering meth- 38. Ripley BD (2007) Pattern recognition and neural networks. Cam-
ods: a case study in Saeen Slope, Azerbaijan province. Iran Catena bridge University Press
120:149–162 39. Ghiasi M, Askarnejad N, Dindarloo SR, Shamsoddini H (2016) Pre-
18. Verma AK, Singh TN, Chauhan NK, Sarkar K (2016) A hybrid diction of blast boulders in open pit mines via multiple regression
FEM–ANN approach for slope instability prediction. J Inst Eng and artificial neural networks. Int J Min Sci Technol 26(2):183–186
(India) 97(3):171–180 40. McClelland JL, Rumelhart DE, PDP Research Group (1986) Parallel
19. Erzin Y, Cetin T (2013) The prediction of the critical factor of safety distributed processing. Explor Microstruct Cogn 2:216–271
of homogeneous finite slopes using neural networks and multiple 41. Demuth H, Beale M (1998) Neural network toolbox for use with
regressions. Comput Geosci 51:305–313 MATLAB. The Math Works Inc., Natick, pp 10–30
20. Pradhan B, Buchroithner MF (2010) Comparison and validation 42. Yılmaz I, Yuksek AG (2008) An example of artificial neural network
of landslide susceptibility maps using an artificial neural net- (ANN) application for indirect estimation of rock parameters. Rock
work model for three test areas in Malaysia. Environ Eng Geosci Mech Rock Eng 41(5):781–795
16(2):107–126 43. Teymen A, Mengüç EC (2020) Comparative evaluation of different
21. Pradhan B, Saied P (2010) Comparison between prediction capa- statistical tools for the prediction of uniaxial compressive strength
bilities of neural network and fuzzy logic techniques for landslide of rocks. Int J Min Sci Technol 30:785
susceptibility mapping. Disaster Adv 3:26–34 44. Garson GD (1991) Interpreting neural-network connection weights.
22. Ray A, Kumar V, Kumar A, Rai R, Khandelwal M, Singh TN (2020) AI Expert 6(7):47–51
Stability prediction of Himalayan residual soil slope using artificial 45. Baruah D, Baruah DC, Hazarika MK (2017) Artificial neural net-
neural network. Nat Hazards 103(3):3523–3540 work based modeling of biomass gasification in fixed bed downdraft
23. Rahul, Khandelwal M, Rai R et al (2015) Evaluation of dump slope gasifiers. Biomass Bioenerg 98:264–271
stability of a coal mine using artificial neural network. Geomech 46. Soofastaei A, Aminossadati SM, Arefi MM, Kizil MS (2016) Devel-
Geophys Geo-energy Geo-resour 1:69–77. https​://doi.org/10.1007/ opment of a multi-layer perceptron artificial neural network model
s4094​8-015-0009-8 to determine haul trucks energy consumption. Int J Min Sci Technol
24. Chakraborty A, Goswami D (2017) Prediction of slope stability 26(2):285–293
using multiple linear regression (MLR) and artificial neural network 47. Erzin Y (2007) Artificial neural networks approach for swell pres-
(ANN). Arab J Geosci 10(17):385 sure versus soil suction behaviour. Can Geotech J 44(10):1215–1223
25. Tien Bui D, Moayedi H, Gör M, Jaafari A, Foong LK (2019) Pre- 48. Erzin Y, Rao BH, Singh DN (2008) Artificial neural network
dicting slope stability failure through machine learning paradigms. models for predicting soil thermal resistivity. Int J Therm Sci
ISPRS Int J Geo-Inf 8(9):395 47(10):1347–1358
26. Quraishi MZ, Mouazen AM (2013) Development of a methodology 49. Erzin Y, Rao BH, Patel A, Gumaste SD, Singh DN (2010) Artificial
for in situ assessment of topsoil dry bulk density. Soil Tillage Res neural network models for predicting electrical resistivity of soils
126:229–237 from their thermal resistivity. Int J Therm Sci 49(1):118–130
27. Mandal U, Gowda V, Ghosh A, Bose A, Bhaumik U, Chatterjee B,
Pal TK (2008) Optimization of metformin HCl 500 mg sustained Publisher’s Note Springer Nature remains neutral with regard to
release matrix tablets using Artificial Neural Network (ANN) jurisdictional claims in published maps and institutional affiliations.
based on Multilayer Perceptrons (MLP) model. Chem Pharm Bull
56(2):150–155
28. Baziar MH, Ghorbani A (2005) Evaluation of lateral spreading using
artificial neural networks. Soil Dyn Earthq Eng 25(1):1–9

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:

1. use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
2. use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
3. falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
4. use bots or other automated methods to access the content or redirect messages
5. override any security feature or exclusionary protocol; or
6. share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at

[email protected]