Professional Documents
Culture Documents
https://doi.org/10.1007/s10706-023-02472-y
ORIGINAL PAPER
Received: 4 July 2022 / Accepted: 29 April 2023 / Published online: 13 May 2023
© The Author(s), under exclusive licence to Springer Nature Switzerland AG 2023
Abstract Predicting the bearing capacity is one of (MAE), root mean square error (RMSE), mean, a20-
the tasks that geotechnical engineers do on a daily index, and coefficient of determination ( R2). The new
basis, yet the accuracy of the available methods needs model scored MAE, RMSE, mean, a20-index, and R2
to be further improved. This paper proposes a new of 65 kPa, 99 kPa, 0.99, 0.68, and 0.97, respectively,
model to accurately predict the bearing capacity of for the training data and 58 kPa, 122 kPa, 0.99, 0.63,
foundations resting on cohesionless soil. The new and 0.97, respectively, for the testing data. The accu-
model has been proposed using a data-driven method racy of the new model has also been compared with
called multi-objective genetic algorithm evolutionary the classical bearing capacity equations of Terzaghi,
computing analysis. The database used in the model Vesic, and Hansen and with other data-driven mod-
development has been collected from previous stud- els, where it was found that the accuracy of the new
ies, and part of this database has been used to test the model is better as it scored better statistical indicators
model to check its accuracy using data that did not and also scored better in the error level-cumulative
influence the model training. The accuracy of the frequency relationship. Thus, the new model can opti-
model has been assessed using mean absolute error mize future designs, as its accuracy has been demon-
strated. Also, this model can be further improved in
the future when new data becomes available.
S. Alzabeebee (*)
Department of Roads and Transport Engineering,
Keywords Bearing capacity · Shallow foundations ·
University of Al-Qadisiyah, Al Diwaniyah, Iraq
e-mail: Saif.Alzabeebee@gmail.com; Saif.Alzabeebee@ EPR-MOGA · Statistical analysis · Cohesionless soil ·
qu.edu.iq Model
Y. M. A. Alshkane
Department of Civil Engineering, College of Engineering,
University of Sulaimani, Sulaimani, Kurdistan Region, 1 Introduction
Iraq
e-mail: younis.ali@univsul.edu.iq The bearing capacity is defined as the stress that the
ground can carry without failure. The bearing capac-
S. Keawsawasvong
Research Unit in Sciences and Innovative Technologies ity of the foundation depends on the dimensions of
for Civil Engineering Infrastructures, Department the foundation, as these will control the overall stress
of Civil Engineering, Thammasat School of Engineering, that is transmitted to the ground. Predicting the bear-
Thammasat University, Khlong Luang 12120,
ing capacity is one of the crucial requirements in the
Pathumthani, Thailand
e-mail: ksurapar@engr.tu.ac.th design. However, practitioners usually calculate the
Vol.: (0123456789)
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
3532 Geotech Geol Eng (2023) 41:3531–3547
bearing capacity using the classical solution (equa- bearing capacity using ANNs. Kohestani et al. (2017)
tion) of Terzaghi (1943) or its later enhancements by compared the predictions of the random forest (RF)
Vesic (1973) or Hansen (1970). Nonetheless, there technique with the classical bearing capacity equa-
are assumptions in the derivation of the original bear- tions and found that RF predicted the bearing capac-
ing capacity equation that warrant the need to exam- ity with better accuracy. Jabbar et al. (2018) estimated
ine the accuracy of this equation and its enhancement, the bearing capacity of shallow foundations using the
as well as find an alternative approach. Hence, there k-nearest neighbor approach (k-nn) and multiple lin-
have been many papers in the literature that utilized ear regression analysis (MLR). They found that the
data-driven methods as an attempt to overcome k-nn approach predicted the bearing capacity with
the limitations of the classical solutions, as will be better accuracy than the MLR analysis.
explained in the next two paragraphs. Omar et al. (2018) utilized ANNs to predict the
Padmini et al. (2008) compared the performance of bearing capacity and settlement from the number of
artificial neural networks (ANNs), fizzy, neurofuzzy, blows of the standard penetration test. The accuracy
and classical bearing capacity solutions in predicting of the ANNs has also been compared with the best
the bearing capacity of shallow foundations resting fit regression analysis, and the results showed that
on cohesionless soils. They noticed that the neuro- the ANNs predicted the bearing capacity and settle-
fuzzy method remarkably outperforms the ANNs and ment better than the best fit regression correlations.
fizzy models. In addition, they also noted that all the Bagińska and Srokosz (2019) examined the capabili-
data-driven methods used in the analysis provided ties of deep neural networks (DNNs) to provide accu-
better estimation of the bearing capacity compared rate predictions of the bearing capacity of shallow
with the classical bearing capacity equations. Zhao foundations. Xue and Chen (2019) built a hybrid least
and Yin (2010) compared the prediction capabilities squares support vector machine (LSSVM) using the
of the chaotic particle swarm optimization-support improved particle swarm optimization (IPSO) algo-
vector machine (CPSO-SVM) with the classical rithm to predict the bearing capacity of shallow foun-
bearing capacity equations for the case of granular dations resting on granular soils. The results of the
soils and noticed that the CPSO-SVM predicted the developed algorithm have been compared with other
bearing capacity with much better accuracy. Adarsh soft computing techniques (fuzzy, neurofuzzy, and
et al. (2012) used support vector machines (SVMs) back-propagation (BP) neural network models) and
and genetic programming (GP) to predict the bear- classical bearing capacity solutions. The results dem-
ing capacity of shallow foundations resting on cohe- onstrated the superiority of the developed algorithm
sionless soils and compared the results with classical over other available solutions/algorithms. However,
bearing capacity equations and other soft computing no symbolic model has been proposed by the authors
results published in the literature (artificial neural net- in this research. Khorrami et al. (2020) developed a
works (ANNs), and fuzzy inference system (FIS)). new model to estimate the bearing capacity of shal-
They noted that the GP achieved better accuracy com- low foundations resting on cohesionless soils utiliz-
pared to the aforementioned techniques. Shahnaz- ing the M5′ model tree (M5′ MD). The results of the
ari and Tutunchian (2012) developed two models to developed model have also been checked against clas-
estimate the bearing capacity of shallow foundations sical bearing capacity equations and the published
using classical multiple linear regression analysis results of other soft computing techniques. They
(MLR) and GP. The accuracy of these models has found that the accuracy of the new model was better
been compared with classical bearing capacity equa- than the classical equations. Also, they demonstrated
tions. Gupta et al. (2016) utilized ANNs to know the that the developed model provided reasonable accu-
minimum number of variables (factors) that need to racy compared with other soft computing results.
be utilized to accurately predict the bearing capacity Based on the aforementioned discussion, it can be
of shallow foundations resting on granular soil. They concluded that although many studies have utilized
found that it is only required to know the bulk unit different data-driven techniques to predict the bear-
weight of the soil, coefficient of curvature and coef- ing capacity of shallow foundations, there have been
ficient of uniformity, footing dimensions, and settle- limited attempts to propose a correlation that can be
ment limit to achieve an accurate prediction of the readily used by practitioners or in future numerical
Vol:. (1234567890)
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Geotech Geol Eng (2023) 41:3531–3547 3533
evaluations by other researchers. Also, the data-driven weight of the soil based on the water table condition
methods used in previous studies are ANNs, CPSO- (γ’ and γ) and the ultimate bearing capacity (qu). The
SVM, SVMs, GP, FIS, RF, k-nn, DNNs, DNNs, and tests reported by Muhs et al. (1969), Weiß (1970),
M5′ MD. However, no study reported the perfor- Muhs and Weiß (1971), and Muhs and Weiß (1973)
mance of the multi-objective evolutionary polynomial were carried out in a test pit in Berlin using a uniform
regression computing (EPR-MOGA), bearing in mind sand-gravel mixture with a mean grain size (D50) of
that this technique has been proven to provide simple 0.65 mm and a coefficient of uniformity of 5.7. The
and robust symbolic models that can be easily used in tests of Briaud and Gibbens (1999) were carried out
the future (Nassr et al. 2018; Jin et al. 2019; Du et al. at an experimental site at the Texas A&M University
2021). Furthermore, the majority of past studies did Riverside Campus. A silty fine silica sand density of
not comprehensively examine the performance of the 0.2 mm and a D 50 of 0.2 mm were used in the tests.
classical bearing capacity solutions. Thus, to fill these Also, the tests reported by Gandhi (2003) were con-
gaps, the objectives of this work are: ducted using a 1-g small-scale model in the Indore
Shri G.S. Institute of Technology and Science in India
1. Provide an extensive statistical examination of using medium, medium dense, and dense sandy soil.
the classical bearing capacity equations. It is necessary to note that the ultimate bearing
2. Examine the capabilities of the EPR-MOGA in capacity in the aforementioned studies was obtained
predicting the bearing capacity of shallow foun- from the plate load test. Also, it is important to point
dations resting on cohesionless soils. out that the effects of particle size distribution and
3. Compare the predictions of the EPR-MOGA with particle shape of soil grains are implicitly included,
classical bearing capacity equations and available as these affect the mobilized angle of internal fric-
symbolic models. tion, which has been used in the model development
of this study. Table 1 shows the statistics (minimum,
maximum, range, average, standard deviation, and
2 Database variance) of the collected database, while Fig. 1 illus-
trates the frequency of each variable.
A database of real field tests is needed to address the
objectives of the current work. Thus, an extensive
search has been conducted to collect useful results
from the literature. This search results in collecting of 3 Available Classical Equations to Predict
97 real laboratory and field tests of the ultimate bear- the Bearing Capacity
ing capacity of shallow foundations resting on cohe-
sionless soils (Muhs et al. 1969; Weiß, 1970; Muhs The bearing capacity can be calculated using equa-
and Weiß, 1971, 1973; Briaud and Gibbens 1999; tions that have been developed based on the failure
Gandhi 2003). This database was originally com- mechanism proposed by Prandtl (1921), the simplest
piled by Padmini et al. (2008). The database contains of which is Terzaghi’s (1943) bearing capacity equa-
the foundation width (B), foundation embedment tion, which is shown in Eq. 1. Also, there are some
(D), foundation length to width ratio (L/B), angle of improvements to this equation proposed by Vesic and
internal friction of the soil (ø), effective or dry unit Hansen, as shown in Eqs. 2 and 3, respectively.
Table 1 Statistics of the Statistical indicator B (m) D (m) L/B γ′ or γ (kN/m3) ø (˚) qu (kPa)
database used in this study
Minimum 0.06 0.00 1.00 9.85 32.0 58.5
Maximum 3.02 0.89 6.00 17.10 44.8 2847.0
Range 2.96 0.89 5.00 7.25 12.8 2788.5
Average 0.40 0.17 3.10 14.18 38.6 439.6
Standard deviation 0.51 0.20 2.14 2.62 3.3 530.9
Variance 0.26 0.04 4.53 6.81 10.6 278,926.9
Vol.: (0123456789)
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
3534 Geotech Geol Eng (2023) 41:3531–3547
Fig. 1 Frequency histograms of the database used in the performance assessment and the evolutionary computing
Vol:. (1234567890)
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Geotech Geol Eng (2023) 41:3531–3547 3535
( )
qu = c� Nc Sc + qNq + 0.5B𝛾 or 𝛾 � N𝛾 S𝛾 (1) 4 Available Data‑Driven Models to Predict
the Bearing Capacity
Vol.: (0123456789)
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
3536 Geotech Geol Eng (2023) 41:3531–3547
Vol:. (1234567890)
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Geotech Geol Eng (2023) 41:3531–3547 3537
Vol.: (0123456789)
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
3538 Geotech Geol Eng (2023) 41:3531–3547
Fig. 3 Comparison of
the no-error line with the
measured-predicted ulti-
mate bearing capacity (pre-
dictions using the Terzaghi
equation with Terzaghi
bearing capacity factors)
Fig. 4 Comparison of
the no-error line with
the measured-predicted
ultimate bearing capac-
ity (predictions using the
Terzaghi equation with
modified Terzaghi bearing
capacity factors)
Fig. 5 Comparison of
the no-error line with
the measured-predicted
ultimate bearing capacity
(predictions using the Vesic
equation)
Vol:. (1234567890)
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Geotech Geol Eng (2023) 41:3531–3547 3539
Fig. 6 Comparison of
the no-error line with
the measured-predicted
ultimate bearing capac-
ity (predictions using the
Hansen equation)
between the predicted and measured values has points it scored R2 of 0.88, which is marginally higher than
close to the error line of − 50%, although Hansen also that scored using the Terzaghi equation, as this equa-
proposed additional factors that ought to enhance the tion scored higher MAE and RMSE and a lower
prediction of the bearing capacity. mean. The Hansen equation also scored a20-index
Furthermore, Fig. 7 compares the MAE, RMSE, of 0.4 (lower than the Terzaghi and Vesic equations),
mean, a20-index, and R 2 of the classical bearing which means only 40% of the predictions were within
capacity equations to gain better insight into the per- an error range of 20%.
formance of these equations. The results of the sta- In summary, the Terzaghi and Vesic equations can
tistical indicators support the initial observations dis- be both ranked first as the difference between them is
cussed in the previous paragraph, where the Terzaghi marginal; the Hansen equation ranks second; and the
and Vesic equations scored better than the other equa- Terzaghi equation using Terzaghi modified bearing
tions. Also, the Terzaghi and Vesic equations scored capacity factors scores last and is not recommended
very comparable performances with a slight differ- to be used in practice.
ence in the MAE, RMSE, mean, a20-index, and R 2.
However, while the Vesic equation scored slightly
higher mean and R2 and lower RMSE, the Terzaghi 8 The Development and the Performance
equation scored lower MAE and a higher a20-index of the EPR‑MOGA Model
(0.59 compared to 0.54 for the Vesic equation), which
means that the Terzaghi equation predicted 59% of The development and performance of the new
data with an error range of no more than 20% com- model are discussed in this section. The collected
pared with 54% of data for the Vesic equation. Fig- database, which has been presented in Fig. 1 and
ure 7 also reveals that while using the Terzaghi equa- Table 1, has been subdivided into two groups: the
tion with modified Terzaghi bearing capacity factors training group and the testing group. The random
provides R2 value of 0.84, these factors scored the function available in Excel has been used to ensure
highest MAE and RMSE, and the lowest mean. The random separation of the data. However, efforts
Terzaghi equation with modified Terzaghi bearing have been made (i.e., a trial and error process) to
capacity factors also scored zero a20-index, indicat- prepare the two groups so that the minimum and
ing that no prediction within acceptable error can be maximum values of the database can be included in
made using these factors. Furthermore, the obtained the training of the EPR-MOGA model to ensure that
indicators show that the Hansen equation performed there are no extrapolations when testing the model
poorly compared with Terzaghi and Vesic, although using the testing group (Alzabeebee et al. 2018;
Vol.: (0123456789)
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
3540 Geotech Geol Eng (2023) 41:3531–3547
Fig. 7 Performance of the classical solutions using different bearing capacity factors
2019). The data preparation technique is similar to 3 show the statistics of the training and testing data
that used in many previous studies (e.g., Alani and groups.
Faramarzi 2014; Ahangar-Asr et al. 2014; Hussain The training and testing groups have been fed into
et al. 2015; Ahangar Asr et al. 2018; Ahangar Asr the EPR-MOGA, and several trials have been under-
and Javadi 2020; Alzabeebee 2022b). Tables 2 and taken to ensure the development of the most accurate
Vol:. (1234567890)
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Geotech Geol Eng (2023) 41:3531–3547 3541
√
EPR-MOGA model. Consequently, the effect of the qu = 4.11≇2 + 1309.98𝛾 − 217.87𝛾 ≇ + 0.103D2 𝛾 1.5 ≇2
selected range of the model exponents, number of �
L (13)
terms of the model, and the model structure have been − 2865.78D2
B
+ 2.159BD≇2 − 3148.4BD1.5 − 5324.65
Table 2 Statistics of the Statistical indicator B (m) D (m) L/B γ′ or γ (kN/m3) ø (°) qu (kPa)
training data group
Minimum 0.06 0.00 1.00 9.85 32.0 58.5
Maximum 3.02 0.89 6.00 17.10 44.8 2847.0
Range 2.96 0.89 5 7.25 12.8 2788.5
Average 0.43 0.19 3.06 14.13 38.5 479.6
Standard deviation 0.55 0.21 2.11 2.65 3.3 564.9
Variance 0.30 0.04 4.45 7.01 10.9 319,179.2
Table 3 Statistics of the Statistical indicator B (m) D (m) L/B γ′ or γ (kN/m3) ø (°) qu (kPa)
testing data group
Minimum 0.09 0.00 1.00 10.20 34.0 91.2
Maximum 0.60 0.15 6.00 17.10 44.8 860.0
Range 0.51 0.15 5.00 6.90 10.8 768.8
Average 0.27 0.05 3.30 14.52 39.0 245.5
Standard deviation 0.20 0.05 2.35 2.54 3.2 230.8
Variance 0.04 0.00 5.51 6.47 10.4 53,286.7
Fig. 8 Comparison of
the no-error line with
the measured-predicted
ultimate bearing capacity
(predictions using the EPR-
MOGA model)
Vol.: (0123456789)
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
3542 Geotech Geol Eng (2023) 41:3531–3547
and testing) are within the error range of ± 20% and an indication of the excellent performance of the
close to the no-error line. These observations give model. Furthermore, Fig. 9 depicts the statistical
Vol:. (1234567890)
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Geotech Geol Eng (2023) 41:3531–3547 3543
measures (MAE, RMSE, mean, a20-index, and R 2) 9 Comparison with Other Data‑Driven Models
of the new model for both data groups. It is evident
from the figure in comparison with the performance In this section, the predictions of the EPR-MOGA
of classical equations presented in Fig. 7 that the symbolic model (Eq. 13) have been compared with
model predicts the ultimate bearing capacity with other symbolic models available in the literature
lower error (MAE equal to 65 kPa for the training (presented in Sect. 4). As discussed previously, these
data group and 58 for the testing data group and models are the multilinear regression model (MLR)
RMSE equal to 99 kPa and 122 kPa for the train- developed by Shahnazari and Tutunchian (2012)
ing and testing data groups, respectively), and a (Eq. 4), the GP model developed by the same afore-
higher R2 (0.97 for both data groups). However, the mentioned authors (Eq. 5), and the M5′ model-tress
scored mean values are similar to those obtained symbolic model (M5′ MT) developed by Khorrami
using the Vesic equation. Importantly, the superior- et al. (2020) (Eq. 6). It is important to stress that
ity of the model can be unarguably noticed in the these models (MLR, GP, and M5′ MT) have also been
scored a20-index in Fig. 9, where the model pre- developed using the same database, and thus, com-
dicted 68% of the training data group and 63% of paring the predictions of these models with the EPR-
the testing data group with an error of prediction of MOGA is fair and meaningful.
not more than ± 20%, which is much better than any The MAE, RMSE, mean, a20-index, and R2 have
of the classical equations. Thus, it can be said based been calculated using the predictions of these models
on the obtained performance that the EPR-MOGA for both the training data group and the testing data
learned to predict the ultimate bearing capacity suc- group and compared with the EPR-MOGA model.
cessfully and provided a robust symbolic model that Table 4 presents the results of the comparisons. It is
offers better prediction than the classical bearing evident from the table that the EPR-MOGA model
capacity equations. beats the other models for all the statistical meas-
Finally, it is important to state that the devel- ures for the training data set, where the EPR-MOGA
oped model is limited by the data used in its train- model scored lower MAE and RMSE, and higher
ing and testing and should not be used in predictions a20-index and R 2. The GP model of Shahnazari and
for scenarios outside the training and testing range Tutunchian (2012) scored second for the training data
(Tables 2, 3). This is indeed one of the shortcomings set as the model obtained better values for the MAE,
of data-driven models. However, the explicit formula- RMSE, a20-index, and R2 compared with other mod-
tion of the model developed in the present study has els. Regarding the testing data, the EPR-MOGA
been provided in Eq. 13, and hence, the model could model also performed better, although it did not pro-
be tested and potentially improved when additional vide the lowest MAE and RMSE, as it scored a higher
reliable data becomes available. a20-index (which is more important) and R2. The
EPR-MOGA model also scored a mean value closer
to the optimum compared with other models for the
testing data group.
Table 4 Performance of the EPR-MOGA model compared with other data-driven models from the literature
Model MAE (kPa) RMSE (kPa) Mean a20-index R2
Vol.: (0123456789)
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
3544 Geotech Geol Eng (2023) 41:3531–3547
Vol:. (1234567890)
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Geotech Geol Eng (2023) 41:3531–3547 3545
6. Thus, the new EPR-MOGA can be considered a attack. Appl Soft Comput 24:985–993. https://doi.org/10.
robust model that can be used to predict the bear- 1016/j.asoc.2014.08.044
Alkroosh I, Alzabeebee S, Al-Taie AJ (2020) Evaluation of
ing capacity of future designs, but within the data the accuracy of commonly used empirical correlations
range used to develop and examine this model. in predicting the compression index of Iraqi fine-grained
However, users should be aware that our pro- soils. Innov Infrastruct Solut 5:68. https://doi.org/10.1007/
posed model can be used in predictions of bear- s41062-020-00321-y
Alzabeebee S (2020) Dynamic response and design of a skirted
ing capacity problems that have parameters lim- strip foundation subjected to vertical vibration. Geomech
ited by the maximum and minimum values of the Eng 20:345–358. https://doi.org/10.12989/gae.2020.20.4.
data used in its training and testing ranges. 345
Alzabeebee S (2022a) Application of EPR-MOGA in comput-
ing the liquefaction-induced settlement of a building sub-
jected to seismic shake. Eng Comput 38:437–448. https://
Acknowledgements This work was supported by the Thai- doi.org/10.1007/s00366-020-01159-9
land Science Research and Innovation Fundamental Fund fiscal Alzabeebee S (2022b) Explicit soft computing model to predict
year 2023. the undrained bearing capacity of footing resting on aggre-
gate pier reinforced cohesive ground. Innov Infrastruct
Author contributions SA: conceptualization, methodology, Solut 7:105. https://doi.org/10.1007/s41062-021-00706-7
validation, formal analysis, writing—original draft. DAM: Alzabeebee S, Chapman DN (2020) Evolutionary computing
data, experimental testing, methodology, writing—review and to determine the skin friction capacity of piles embedded
editing. YMA: methodology, writing—review and editing. in clay and evaluation of the available analytical methods.
Transport Geotech 24:100372. https://doi.org/10.1016/j.
Funding No funding was received for conducting this study. trgeo.2020.100372
Alzabeebee S, Chapman DN, Faramarzi A (2018) Develop-
Availability of Data and Materials Data used in this ment of a novel model to estimate bedding factors to
research are available upon request. ensure the economic and robust design of rigid pipes
under soil loads. Tunnel Undergr Space Techno 71:567–
Code Availability Code used in this research is available 578. https://doi.org/10.1016/j.tust.2017.11.009
upon request. Alzabeebee S, Chapman DN, Faramarzi A (2019) Economical
design of buried concrete pipes subjected to UK standard
Declarations traffic loading. Proc Inst Civ Eng Struct Build 172:141–
156. https://doi.org/10.1680/jstbu.17.00035
Conflict of interest The authors declare no conflict of interest Alzabeebee S, Zuhaira AA, Al-Hamd RKS (2022) Develop-
associated with this submission. In addition, the authors have no ment of an optimized model to compute the undrained
relevant financial or non-financial interests to disclose. shaft friction adhesion factor of bored piles. Geomech Eng
28:397–404. https://doi.org/10.12989/gae.2022.28.4.397
Armaghani DJ, Mamou A, Maraveas C, Roussis PC, Siorikis
VG, Skentou AD, Asteris PG (2021) Predicting the
unconfined compressive strength of granite using only two
References non-destructive test indexes. Geomech Eng 25:317–330.
https://doi.org/10.12989/gae.2021.25.4.317
Adarsh S, Dhanya R, Krishna G, Merlin R, Tina J (2012) Pre- Bagińska M, Srokosz PE (2019) The optimal ANN Model for
diction of ultimate bearing capacity of cohesionless soils predicting bearing capacity of shallow foundations trained
using soft computing techniques. Int Sch Res Not. https:// on scarce data. KSCE J Civ Eng 23:130–137. https://doi.
doi.org/10.5402/2012/628496 org/10.1007/s12205-018-2636-4
Ahangar Asr A, Javadi AA (2020) Effective stress parameter in Bai XD, Cheng WC, Ong DE, Li G (2021) Evaluation of
unsaturated soils: an evolutionary-based prediction model. geological conditions and clogging of tunneling using
Proc Inst Civ Eng Smart Infrastruct Constr 173:96–105. machine learning. Geomech Eng 25(1):59–73. https://doi.
https://doi.org/10.1680/jsmic.21.00012 org/10.12989/gae.2021.25.1.059
Ahangar Asr A, Faramarzi A, Javadi AA (2018) An evolution- Bowles JE (1996) Foundation analysis and design. McGraw-
ary modelling approach to predicting stress–strain behav- Hill, New York
iour of saturated granular soils. Eng Comput 35:2931– Briaud JL, Gibbens R (1999) Behavior of five large spread
2952. https://doi.org/10.1108/EC-01-2018-0025 footings in sand. J Geotech Geoenviron Eng 125:787–796.
Ahangar-Asr A, Javadi AA, Johari A, Chen Y (2014) Lat- https:// d oi. o rg/ 1 0. 1 061/ ( ASCE) 1 090- 0 241(1999) 1 25:
eral load bearing capacity modelling of piles in cohesive 9(787)
soils in undrained conditions: an intelligent evolutionary Das BM (2007) Principles of foundation engineering. Cengage
approach. Appl Soft Comput 24:822–828. https://doi.org/ Learning, Boston
10.1016/j.asoc.2014.07.027 Du Z, Shahin MA, El Naggar H (2021) Design of ram-
Alani AM, Faramarzi A (2014) An evolutionary approach to compacted bearing base piling foundations by simple
modelling concrete degradation due to sulphuric acid numerical modelling approach and artificial intelligence
Vol.: (0123456789)
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
3546 Geotech Geol Eng (2023) 41:3531–3547
technique. Int J Geosynth Ground Eng 7:41. https://doi. conference on soil mechanism and foundation engineer-
org/10.1007/s40891-021-00287-6 ing, II, pp 173–179
Gandhi GN (2003) Study of bearing capacity factors developed Muhs H, Elmiger R, Weiß K (1969) Sohlreibung und
from lab. Experiments on shallow footings on cohesion- Grenztragfӓhigkeit unter lotrecht und schrӓg belasteten
less soils. Ph.D. thesis, Shri G.S. Institute of Technology Einzelfundamenten. Deutsche Forschungsgesellschaft
and Science für Bodenmechanik (DEGEBO), Berlin. HEFT 62 (in
Giustolisi O, Savic DA (2006) A symbolic data-driven tech- German)
nique based on evolutionary polynomial regression. J Nassr A, Esmaeili-Falak M, Katebi H, Javadi A (2018) A
Hydroinform 8:207–222. https://doi.org/10.2166/hydro. new approach to modeling the behavior of frozen soils.
2006.020b Eng Geol 246:82–90. https://doi.org/10.1016/j.enggeo.
Giustolisi O, Savic DA (2009) Advances in data-driven analy- 2018.09.018
ses and modelling using EPR-MOGA. J Hydroinform Omar M, Hamad K, Al Suwaidi M, Shanableh A (2018)
11:225–236. https://doi.org/10.2166/hydro.2009.017 Developing artificial neural network models to pre-
Gupta R, Goyal K, Yadav N (2016) Prediction of safe bearing dict allowable bearing capacity and elastic settlement
capacity of noncohesive soil in arid zone using artificial of shallow foundation in Sharjah, United Arab Emir-
neural networks. Int J Geomech 16:04015044. https://doi. ates. Arab J Geosci 11:1–11. https://doi.org/10.1007/
org/10.1061/(ASCE)GM.1943-5622.0000514 s12517-018-3828-4
Hansen JB (1970) A revised and extended formula for bearing Padmini D, Ilamparuthi K, Sudheer KP (2008) Ultimate bear-
capacity. Dan Geotech Inst 28:5–11 ing capacity prediction of shallow foundations on cohe-
Hussain MS, Javadi AA, Ahangar-Asr A, Farmani R (2015) sionless soils using neurofuzzy models. Comput Geotech
A surrogate model for simulation–optimization of aqui- 35:33–46. https://doi.org/10.1016/j.compgeo.2007.03.001
fer systems subjected to seawater intrusion. J Hydrol Prandtl L (1921) Über die Eindringungsfestigkeit (Härte) plas-
523:542–554. https://doi.org/10.1016/j.jhydrol.2015.01. tischer Baustoffe und die Festigkeit von Schneiden. Z
079 Angew Math Mech 1(1):15–20 (in German)
Jabbar SF, Hamed RI, Alwan AH (2018) The potential of non- Shahnazari H, Tutunchian MA (2012) Prediction of ultimate
parametric model in foundation bearing capacity predic- bearing capacity of shallow foundations on cohesion-
tion. Neural Comput Appl 30:3235–3241 less soils: An evolutionary approach. KSCE J Civ Eng
Jin YF, Yin ZY, Zhou WH, Yin JH, Shao JF (2019) A single- 16:950–957. https://doi.org/10.1007/s12205-012-1651-0
objective EPR based model for creep index of soft clays Terzaghi K (1943) Theoretical soil mechanics. Wiley, New
considering L2 regularization. Eng Geol 248:242–255. York
https://doi.org/10.1016/j.enggeo.2018.12.006 Uysal F (2020) Prediction of collapse potential of soils
Khorrami R, Derakhshani A, Moayedi H (2020) New explicit using gene expression programming and parametric
formulation for ultimate bearing capacity of shallow foun- study. Arab J Geosci 13:1038. https://doi.org/10.1007/
dations on granular soil using M5’model tree. Measure- s12517-020-06050-x
ment 163:108032. https://doi.org/10.1016/j.measurement. Vesic AS (1973) Analysis of ultimate loads of shallow founda-
2020.108032 tions. J Soil Mech Found Div 99:45–73. https://doi.org/10.
Kohestani VR, Vosoghi M, Hassanlourad M, Fallahnia M 1061/JSFEAQ.0001846
(2017) Bearing capacity of shallow foundations on cohe- Weiß K (1970) Der Einfluß der Fundamentform auf die
sionless soils: a random forest based approach. Civ Eng Grenztragfӓhigkeit flachgegründeter Fundamente.
Infrastruct J 50:35–49. https://doi.org/10.7508/ceij.2017. Deutsche Forschungsgesellschaft für Bodenmechanik
01.003 (DEGEBO), Berlin. HEFT 65 (in German)
Kumbhojkar AS (1993) Numerical evaluation of Terzaghi’s N Xue X, Chen X (2019) Determination of ultimate bearing
γ. J Geotech Eng 119:598–607. https://doi.org/10.1061/ capacity of shallow foundations using LSSVM algorithm.
(ASCE)0733-9410(1993)119:3(598) J Civ Eng Manag 25:451–459. https://doi.org/10.3846/
Luat NV, Lee K, Thai DK (2020a) Application of artificial neu- jcem.2019.9875
ral networks in settlement prediction of shallow founda- Zhang WG, Goh ATC (2013) Multivariate adaptive regression
tions on sandy soils. Geomech Eng 20:385–397. https:// splines for analysis of geotechnical engineering systems.
doi.org/10.12989/gae.2020.20.5.385 Comput Geotech 48:82–95. https://doi.org/10.1016/j.
Luat NV, Nguyen VQ, Lee S, Woo S, Lee K (2020b) An evo- compgeo.2012.09.016
lutionary hybrid optimization of MARS model in pre- Zhang W, Goh AT (2016) Multivariate adaptive regression
dicting settlement of shallow foundations on sandy soils. splines and neural network models for prediction of pile
Geomech Eng 21:583–598. https://doi.org/10.12989/gae. drivability. Geosci Front 7:45–52. https://doi.org/10.
2020.21.6.583 1016/j.gsf.2014.10.003
Muhs H, Weiß K (1971) Untersuchung von Grenztragfӓhigkeit Zhang W, Goh AT, Zhang Y, Chen Y, Xiao Y (2015) Assess-
und Setzungsverhalten flachgegründeter Einzelfun- ment of soil liquefaction based on capacity energy concept
damente im ungleichfӧrmigennichtbindigen Boden. and multivariate adaptive regression splines. Eng Geol
Deutsche Forschungsgesellschaft für Bodenmechanik 188:29–37. https://doi.org/10.1016/j.enggeo.2015.01.009
(DEGEBO), Berlin. HEFT 69 (in German) Zhang W, Zhang R, Wu C, Goh ATC, Lacasse S, Liu Z, Liu H
Muhs H, Weiß K (1973) Inclined load tests on shallow (2020) State-of-the-art review of soft computing applica-
strip footings. In: Proceedings of the 8th international tions in underground excavations. Geosci Front 11:1095–
1106. https://doi.org/10.1016/j.gsf.2019.12.003
Vol:. (1234567890)
13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Geotech Geol Eng (2023) 41:3531–3547 3547
Zhao HB, Yin S (2010) A CPSO-SVM model for ultimate steels by regression and machine learning. Multiscale
bearing capacity determination. Mar Georesour Geotech- Multidiscip Model Exp and Des 5:337–350. https://doi.
nol 28:64–75. https://doi.org/10.1080/106411909033590 org/10.1007/s41939-022-00123-y
76
Zhu W, Rad HN, Hasanipanah M (2021) A chaos recurrent Publisher’s Note Springer Nature remains neutral with regard
ANFIS optimized by PSO to predict ground vibration gen- to jurisdictional claims in published maps and institutional
erated in rock blasting. Appl Soft Comput 108:107434. affiliations.
https://doi.org/10.1016/j.asoc.2021.107434
Zhu W, Huang L, Zhang Z (2022a) Novel hybrid AOA and
Springer Nature or its licensor (e.g. a society or other partner)
ALO optimized supervised machine learning approaches
holds exclusive rights to this article under a publishing
to predict the compressive strength of admixed concrete
agreement with the author(s) or other rightsholder(s); author
containing fly ash and micro-silica. Multiscale Multidis-
self-archiving of the accepted manuscript version of this article
cip Model Exp Des 5:391–402. https://doi.org/10.1007/
is solely governed by the terms of such publishing agreement
s41939-022-00124-x
and applicable law.
Zhu XK, Zhu JB, Zhang W (2022b) Data-driven models of
dynamic strength of resistance spot welds in high strength
Vol.: (0123456789)
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
onlineservice@springernature.com