Geoscience Frontiers 8 (2017) 199e204

H O S T E D BY Contents lists available at ScienceDirect

China University of Geosciences (Beijing)

Geoscience Frontiers
journal homepage: www.elsevier.com/locate/gsf

Research paper

Predicting standard penetration test N-value from cone penetration

test data using artificial neural networks
Bashar Tarawneh*
Department of Civil Engineering, The University of Jordan, Amman, 11942, Jordan

a r t i c l e i n f o a b s t r a c t

Article history: Standard Penetration Test (SPT) and Cone Penetration Test (CPT) are the most frequently used field tests
Received 11 July 2015 to estimate soil parameters for geotechnical analysis and design. Numerous soil parameters are related to
Received in revised form the SPT N-value. In contrast, CPT is becoming more popular for site investigation and geotechnical design.
26 January 2016
Correlation of CPT data with SPT N-value is very beneficial since most of the field parameters are related
Accepted 17 February 2016
to SPT N-values. A back-propagation artificial neural network (ANN) model was developed to predict the
Available online 21 March 2016
N60-value from CPT data. Data used in this study consisted of 109 CPT-SPT pairs for sand, sandy silt, and
silty sand soils. The ANN model input variables are: CPT tip resistance (qc), effective vertical stress ðs0v Þ,
Keywords:
SPT
and CPT sleeve friction (fs). A different set of SPT-CPT data was used to check the reliability of the
CPT developed ANN model. It was shown that ANN model either under-predicted the N60-value by 7e16% or
Correlation over-predicted it by 7e20%. It is concluded that back-propagation neural networks is a good tool to
Artificial neural network predict N60-value from CPT data with acceptable accuracy.
Sand Ó 2016, China University of Geosciences (Beijing) and Peking University. Production and hosting by
Silt Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
licenses/by-nc-nd/4.0/).

1. Introduction
For site exploration, in situ tests are used to delineate soil
stratigraphy and determine its properties for geotechnical analysis
and design. The penetration resistances are used to classify and
characterize subsoils. Substantial data can be obtained economi-
cally in shorter time using in situ devices, such as the standard
penetration test (SPT) and cone penetration test (CPT). Some
geotechnical design parameters of the soil are associated with the
SPT. In contrast CPT is becoming more popular for site investigation
and geotechnical design. In construction projects, it is common to
use SPT for the preliminary soil investigation, whereas CPT is used
for detailed soil investigation and construction quality control.
Correlation of cone tip resistance, qc, with SPT N-value is a very
beneficial approach since most field parameters are based on SPT
N-values and CPT tip resistance.
The objective of this paper is to develop an Artificial Neural
Networks (ANNs) model that can predict N-value from CPT data for
cohesionless soils.
* Tel.: þ962 79 754 9696; fax: þ962 6 5355589.
E-mail address: btarawneh@ju.edu.jo.
Peer-review under responsibility of China University of Geosciences (Beijing).
The relationship between SPT N-values and cone tip resistance
qc is defined by a ratio “n” where n ¼ qc/N. De Alencar Velloso
(1959) presented ratios of cone tip resistance to SPT N-value for
different soil types, 0.35 MPa for clay and silty clay, 0.2 MPa for
sandy clay and silty clay, 0.35 MPa for sandy silt, 0.6 MPa for fine
grained sand, and 1.0 MPa for sand. Meigh and Nixon (1961) rec-
ommended n-values as 0.2 MPa for coarse grained sand and
0.3e0.4 MPa for gravelly sand. Schmertmann (1970) proposed n-
value of n ¼ (qc þ fs)/N as 0.2 MPa for silt, sandy silt, and silt-sand
mixture, 0.3e0.4 MPa for fine to medium sand and silty sand,
0.5e0.6 MPa for coarse sand, sand with gravel, and 0.8e1.0 MPa for
sandy gravel and gravel.
Robertson et al. (1983) demonstrated the qc/N ratio as a function
of mean grain size, D50, the geology of the study area where over-
consolidation exists. They proposed a soil behavior-type classifi-
cation zone based on cone penetration test with pore pressure
measurement tests (CPTU, piezocone). Ismael and Jeragh (1986)
correlated CPT qc values with SPT N-values for calcareous desert
sands in Kuwait and compared it with the results of Schmertmann
(1970) for clean, fine to medium sands and slightly silty sands. Their
proposed n-values were higher than those of Schmertmann (1970)
for clean, fine to medium sands and slightly silty sands. A close
agreement of their test results in the form of qc/N versus mean grain
size ‘D50’ where found when compared with the historical data of
Robertson et al. (1983).

http://dx.doi.org/10.1016/j.gsf.2016.02.003
1674-9871/Ó 2016, China University of Geosciences (Beijing) and Peking University. Production and hosting by Elsevier B.V. This is an open access article under the CC BY-NC-
ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
200 B. Tarawneh / Geoscience Frontiers 8 (2017) 199e204
Jefferies and Davies (1993) proposed a soil classification chart
estimating N-values. This chart considered qc by taking into account
pore water pressure (u) and overburden effective stress ðs0vo Þ using
piezocone.
Danziger and de Velloso (1995) developed a correlation be-
tween CPT and SPT for Brazilian soils. Values found in the same
range were obtained by Schmertmann (1970). Different types of
correlation were tested, and a linear correlation was found better
suited for practical applications. A general trend was obtained
similar to Robertson’s curve (increasing n-values with increasing
grain size).
Akca (2003) presented SPT-CPT correlation for United Arab
Emirates soils. Results of his study showed higher values of n ¼ qc/N
when compared to those found in the literature. He explained that
higher values are due to cementation, densification and Shelly
structure or gravel layers in the United Arab Emirates soils. Shahri
et al. (2014) developed a correlation between qc and N-value for
various soil layers, particularly in clayey soils with significant clay
content in an area in southwest Sweden. They proposed linear and
power relationships to predict qc using N-value. The results of their
study showed a good agreement with previous work by other re-
searchers. Tarawneh (2014) developed a multiple linear regression
(MLR) and a symbolic regression (SR) models to predict N-value
using CPT data for sand, sandy silt, and silty sand soils. Asci et al.
(2015) proposed exponential models to predict CPT qc value from
SPT N-values for silty clay, clayey silt, clay, and sandy silt.
2. Data collection
This research was carried out using existing SPT-CPT pairs
collected in Dubai and Abu-Dhabi, United Arab Emirates. Data used
in this study consisted of 109 CPT-SPT pairs for sand, sandy silt, and
silty sand soils. Distance between each CPT-SPT pair ranged from 2
to 25 m. The depth of the SPT-CPT pairs ranged from 3 to 12 m.
Water table was encountered in all CPT-SPT pairs between 0.8 and
5.8 m below existing ground level. The Unified Soil Classification
System (USCS) was used to classify the collected samples of the
SPTs. Tip resistance (qc) and friction ratio (Rf) of the CPTs were used
to classify the soil based on the soil behavior type (SBT) charts
proposed by Lunne et al. (1997) and Robertson (2009). The results
showed that the soil can be classified as sand, sandy silt, and silty
sand.
Each CPT collects one reading every 0.02 m while the SPT has
one reading every 0.5 m. Therefore, CPT results were averaged over
0.5 m intervals. This average was compared with the SPT N-value
located over the same depth range.
It is essential to normalize the N-values measured by any
hammer to a standard rod energy ratio. The N-value in a given soil
is inversely proportional to rod energy ratio (ERr) (Skempton,
1986). N-values measured with a known or estimated rod energy
ratio (ERr) value can be normalized to this standard using N60 ¼ N
(ERr/60). Skempton (1986) gave the rod energy ratio for different
hammer types and release systems. Therefore, all N-values were
normalized to 60% hammer efficiency to calculate N60. Table 1
Table 1
Collected data statistics.
Variable N60 Tip Sleeve Friction Effective
resistance friction fs ratio Rf (%) stress s0v
qc (MPa) (MPa) (kPa)
Mean 12.2 9.4 0.078 0.64 42.1
Min. 3 0.24 0.001 0.08 0.4
Max. 50 38.3 0.66 1.93 135.2
Range 47 38.06 0.659 1.85 134.8
summarizes the mean, minimum, maximum, range of the
collected field data, the calculated friction ratio Rf ¼ (fs/qc)  100%,
and the effective vertical stress.
3. Artificial neural networks (ANNs)
Artificial neural networks (ANNs) are a form of Artificial Intel-
ligence (AI) that attempt to mimic the function of the human brain
and nervous system. Over the last decade, artificial intelligence (AI)
has been applied successfully to many problems in geotechnical
engineering. Examples of the available AI techniques are artificial
neural networks (ANNs), evolutionary polynomial regression (EPR),
genetic programming (GP), M5 model trees, support vector ma-
chines (SVM), and k-nearest neighbors (Elshorbagy et al., 2010). Of
these, ANNs are by far the most commonly used AI technique in
geotechnical engineering. More recently, GP and EPR have been
frequently used in geotechnical engineering and have proved to be
successful (Shahin, 2014).
Bendana et al. (2008) described ANN as “massively parallel
distributed processor” which can store information taken from a
data set that is supplied out of the network. The ANN system
consists of three or more layers. The first layer has the input neu-
rons (parameters), while the last layer contains the output. In be-
tween are one or more hidden layers, which are for delineating and
learning the patterns governing the network’s data. The develop-
ment of an ANN model requires the determination of model inputs
and outputs, division and pre-processing of the available data, the
determination of appropriate network architecture, stopping, and
model validations.
Shahin et al. (2009) presented an overview of the ANN’s archi-
tecture. It consists of a series of processing elements (PEs), or nodes,
that are usually arranged in layers: an input layer, an output layer,
and one or more hidden layers, as shown in Fig. 1. The input from
each PE in the previous layer xi is multiplied by an adjustable
connection weight wji. At each PE, the weighted input signals are
summed and a threshold value, qj is added. This combined input, Ij
is then passed through a nonlinear transfer function f(Ij) to produce
the output of the PE yj. The output of one PE provides the input to
the PEs in the next layer. This process is summarized in Eqs. (1) and
(2) and Fig. 1.
X
Ij ¼ wji xi þ qj (1)
 
yj ¼ f Ij (2)
Previous research in the field of geotechnical engineering made
use of ANN models. Goh (1995) developed a neural network model
to provide initial estimates of maximum wall deflections for braced
excavations in soft clay. Ni et al. (1996) proposed a methodology of
combining fuzzy sets theory with artificial neural networks for
evaluating the stability of slopes. Zhou and Wu (1994) used a neural
network model to characterize the spatial distribution of rock head
elevations. Shi et al. (1998) presented a study of neural networks for
predicting settlements of tunnels. Shahin et al. (2000) presented a
back-propagation artificial neural network model for predicting the
settlement of shallow foundations on cohesionless soils. Nejad et al.
(2009) developed an ANN model to predict pile settlement based
Depth of on standard penetration test. Tatari et al. (2013) presented an
collected artificial neural network (ANN) model to assess the condition of
data (m) culverts based on inventory data presented by Al Tarawneh (2005)
4.2 and Masada et al. (2006, 2007). Tarawneh (2013) and Tarawneh and
0.02 Imam (2014) developed ANNs models that can predict pile setup for
12
11.98
three pile types (pipe, concrete, and H-pile) using dynamic load
tests. Tarawneh and Nazzal (2014) employed ANN to optimize the
 B. Tarawneh / Geoscience Frontiers 8 (2017) 199e204 201

Figure 1. General structure and process of ANNs (Shahin et al., 2009).

prediction of subgrade resilient modulus design input from falling
weight deflectometer test results.
3.1. ANNs model to predict N60 from CPT test
In this study, Neuro-Solutions 6.0 Software was used in devel-
oping the neural network models. This software combines a
modular design interface with advanced learning procedures, giv-
ing the power and flexibility needed to design the neural network
that produces the best solution.
The ANN model input variables are CPT tip resistance (qc),
effective vertical stress ðs0v Þ, and CPT sleeve friction (fs). The ANN
model output is SPT N60-value. The data was divided into three sets:
training, cross validation, and testing. 70% of the data points were
selected for training, 15% were selected for cross validation, and 15%
were used for testing the network. The training data points were
used to train the network and compute the weights of the inputs.
The cross validation computes the error in a test set at the same
time that the network is being trained with the training set. The test
data points were used to measure the performance of the selected
ANN model.
It is important that the data used for training, cross validation,
and testing represent the same population and the statistical
properties. In order to develop the best possible model, all patterns
that are contained in the data need to be included in the training
set. Similarly, since the test set is used to determine when to stop
training, it needs to be representative of the training set and should
contain all of the patterns that are present in the available data
(Shahin et al., 2002). To accomplish this, several random combi-
nations of the training, cross validation and testing sets were tried
until a statistically consistent data set was obtained.
3.2. ANN model architecture
A total of three input variables were included in the ANN model
namely, CPT tip resistance (qc), effective vertical stress ðs0v Þ, and CPT
sleeve friction (fs). The output layer has a single node representing
the SPT N60-value. Several network structures, with different
numbers of hidden layers and nodes in the hidden layer, were
trained and tested to find the model with best performing network
architecture. Although it has been shown that a network with one
hidden layer can approximate any continuous function (Hornik
et al., 1989), in this research one and two hidden layers were
employed. In order to determine the optimum network geometry,
first ANNs with a single hidden layer and different number of nodes
in the hidden layer were trained. Then ANNs with two hidden
layers using different number of nodes in the hidden layers were
also trained. Sigmoid (Sig.) and hyperbolic tangent (tanh) transfer
functions for the hidden and output layers were employed. Com-
binations of number of elements in each hidden layer and types of
transfer function that yielded the most accurate predictions of N60-
value are presented in Tables 2 and 3.
3.3. ANN model optimization
ANN models to predict SPT N60-value from CPT data using back-
propagation algorithms are developed. The weights of the network
are tuned during the training phase to minimize the error. In each
iteration, the error propagates backward to minimize the error to a
desired level. The back-propagation algorithm is used for opti-
mizing the connection weights in this study, whereas the
Levenberg-Marquardt (LM) algorithm was used as a learning rule. It
is one of the most appropriate higher-order adaptive algorithms
known for minimizing the error of a neural network (Principe et al.,

Table 2
Single hidden layer ANN models.

Model no. Input nodes Hidden layer-1 Output layer Testing data

Processing elements Transfer function Processing elements Transfer function MAE R

1 3
2 3
3 3
4 3
5 3
6 3
7 3
2 tanh 1 tanh 4.89 0.81
3 tanh 1 tanh 5.89 0.79
4 tanh 1 tanh 4.51 0.9
5 tanh 1 tanh 4.55 0.92
6 tanh 1 tanh 3.15 0.94
6 Sig. 1 Sig. 2.88 0.95
7 tanh 1 tanh 3.86 0.88
202 B. Tarawneh / Geoscience Frontiers 8 (2017) 199e204

Table 3
Two hidden layer ANN models.

Model no. Input nodes Hidden layer-1 Hidden layer-2 Output layer Testing data

Processing Transfer Processing Transfer Processing Transfer MAE R

elements function elements function elements function
8 3 2 Sig. 2 Sig. 1 Sig. 2.92 0.94
9 3 2 tanh 2 tanh 1 tanh 2.94 0.90
10 3 3 Sig. 3 Sig. 1 Sig. 5.04 0.77
11 3 3 tanh 3 tanh 1 tanh 4.04 0.92
12 3 6 Sig. 6 Sig. 1 Sig. 3.77 0.91
13 3 6 tanh 6 tanh 1 tanh 3.74 0.93

1999). The cross-validation technique was used in this research as a
stopping criteria, as it ensures over-fitting does not occur. The
training set was used to adjust the connection weights, while the
testing set measured the capability of the model to generalize.
After achieving a trained model, its performance must be vali-
dated using data sets that have not been used during the learning
process, this data set is called testing set. The purpose of the model
validation stage is to make sure that the model has the ability to
generalize the input-output relations that are contained in the
training data (Shahin et al., 2002). The coefficient of correlation (R),
the root mean squared error (RMSE), and the mean absolute error
(MAE) are the main criteria that are used to assess the prediction
performance of the developed ANN models.
3.4. ANN outcomes
For the one hidden layer models, Table 2 shows the results of the
best performing models. The MAE values ranged from 2.88 to 5.89,
while the coefficient of correlation (R) values were between 0.79 and
0.95 for the testing data set. Model 6 which has six processing ele-
ments in the hidden layer, using sigmoid as transfer function for both
hidden and output layers, was the best performing among the single
hidden layer models. Table 3 shows the results of the best performing
models with two hidden layers. The MAE values ranged from 2.92 to
5.04, while the coefficient of correlation (R) values ranged from 0.77
to 0.94 for the testing data set. Model 8 which has two processing
elements in each hidden layer, using sigmoid as transfer function for
all layers, was the best performing among all the developed two
hidden layer models. To summarize, model 6 is the best performing
ANN model. Based on the available data and results, model 6 is rec-
ommended to predict SPT N60-value from CPT data.
3.5. ANN model 6
Fig. 2 shows the structure of the ANN model 6. In this model I1,
I2, and I3 represent CPT tip resistance (qc), CPT sleeve friction (fs),
and effective vertical stress ðs0v Þ, respectively. This model has one
hidden layer with six nodes in the hidden layer. Tables 4e7 provide
the numerical values of the ANN amplitude, offset, weights and
biases. In order to use the developed ANN model 6 to calculate N60-
value, a four step procedure is provided below:
(1) The first step is to calculate the normalized inputs (IN) for the
inputs I1 through I3 by using the amplitude (ain) and the offset
(ofin) values provided in Table 4 (IN ¼ ainIn þ ofin). Each input is
multiplied by the amplitude and shifted by an offset. The
amplitude and offset are often referred to as normalization
coefficients.
(2) The second step is to calculate inputs and outputs at each node
in the hidden layer (H1, H2, H3, H4, H5, and H6).
X
Input at each node in the hidden layer; IH ¼ winm ,IN þ bhm
(3)
1
Output at each node in the hidden layer; OH ¼ (4)
1 þ eIH
(3) The third step is to calculate the inputs and outputs at the node
in the output layer (O).

Figure 2. Structure of ANN model 6.

 B. Tarawneh / Geoscience Frontiers 8 (2017) 199e204 203

Table 4 4. Verification of the ANNs models

Amplitude and offset for the input layer.

Number of nodes 3 To achieve a higher reliability of the developed ANN models, a

Amplitude, ain Offset, ofin
different set of SPT-CPT data were used to predict N60 -value
using the best performing ANN model 6. Ten data points were
0.024 0.042
1.572 0.050 used to predict N60 -value at different depths ranging from 0.6 to
0.008 0.010 6.0 m. Predicted N60 -values were compared to the field N60 -
value as shown in Table 8. It can be noted that ANN model 6
either under-predicted the N60 -value by 7e16% or over-
predicted it by 7e20%.
Table 5
Weight of inputs to hidden layer (winm)a and biases (bh1 to bh6).
5. Conclusions
wi11 8.7086 wi21 10.2144 wi31 6.6877 bh1 6.5516
wi12 2.2677 wi22 0.7720 wi32 2.4656 bh2 1.4991
wi13 11.7744 wi23 4.1926 wi33 0.1428 bh3 0.3714 The use of ANNs to predict N60-value using CPT data was
wi14 8.6363 wi24 6.7383 wi34 14.2027 bh4 9.0208 assessed in this paper. A back-propagation neural network was
wi15 15.3545 wi25 1.4264 wi35 9.9004 bh5 2.5551 used to examine the feasibility of ANNs to predict the N60-value.
wi16 9.7038 wi26 1.6743 wi36 12.9872 bh6 8.1755
Data used in this study consisted of 109 CPT-SPT pairs for sand,
a
winm represents the weight from input n to hidden node m. sandy silt, and silty sand soils. The ANN model input variables are
CPT tip resistance (qc), effective vertical stress (sv0 ), and CPT sleeve
friction (fs). The ANN model output is SPT N60-value. Model 6 which
Table 6 has one hidden layer with six processing elements in the hidden
Weights of hidden layer to output layer (wo) and bias of output node (bo).
layer, using sigmoid as transfer function, was the best performing
wo1 1.3138 model among all ANN models. Model 6 indicated that back-
wo2 6.8304 propagation neural networks have the ability to predict the N60-
wo3 5.2317
value with an adequate accuracy (R ¼ 0.95, MAE ¼ 2.88).
wo4 9.6552
wo5 1.6084 As extra step to check the reliability of the developed ANN
wo6 9.6662 model 6, a different set of SPT-CPT data were used to predict N60-
bo 2.69144 value using the ANN model 6. It was shown that ANN model 6
either under-predicted the N60-value by 7e16% or over-predicted it
by 7e20%. It can be concluded that ANN is a good tool to predict
Table 7 N60-value from CPT data with acceptable accuracy.
Amplitude and offset of the output layer.

Number of nodes ¼ 1 Acknowledgment

Amplitude, ao Offset, ofo
The author would like to express his appreciation to Rapid
0.01837 0.03163
Impact Compaction Contracting L.L.C. (RlC), for making the SPT-CPT
database available for this work.

X References
Input at the output layer; Io ¼ woi ,HN þ bhm (5)
Akca, N., 2003. Correlation of SPTeCPT data from the United Arab Emirates. Engi-
neering Geology 67 (3e4), 219e231.
1
Final output ¼ (6) Al Tarawneh, B.K., 2005. Inspection, Durability, and Risk Assessment of Highway
1 þ eIO Culverts. Ph.D. thesis. Ohio University, Athens, Ohio.
De Alencar Velloso, Di, 1959. O ensaio de diepsondeering e a determinacao da
capacidade de carga do solo. Rodovia, 29.
Asci, M., Kurtulus, C., Kaplanvural, I., 2015. Correlation of SPT-CPT data from the
subsidence area in the Golcuk, Turkey. Soil Mechanics and Foundation Engi-
(4) The last step is to de-normalized the output by subtracting the neering 51 (6), 268e272.
offset at the output (ofo) from Eq. (6) then divided the result by Bendana, R., Del Cano, A., De la Cruz, M.P., 2008. Contractor selection: fuzzy-control
the amplitude at the output layer (ao) node. approach. Canadian Journal of Civil Engineering 35 (5), 473e486.
Danziger, B.R., Velloso, D.A., 1995. Correlations between the CPT and the SPT for
some Brazilian soils. In: Proceedings, Cone Penetration Testing (CPT 95), Link-
öping, Sweden, vol. 2, pp. 155e160.
Table 8 Elshorbagy, A., Corzo, G., Srinivasulu, S., Solomatine, D.P., 2010. Experimental
Verification of the developed models using different data set. investigation of the predictive capabilities of data driven modeling techniques
in hydrology-part 1: concepts and methodology. Hydrology and Earth System
CPT data N60-field ANN model 6
Science 14, 1931e1941.
qc (MPa) fs (MPa) s0v (kPa) N60p N60p/N60f Goh, A.T.C., 1995. Empirical design in geotechnics using neural networks. Geo-
technique 45 (4), 709e714.
4 0.015 27.23 7 7.5 1.07 Hornik, K., Stinchcombe, M., White, H., 1989. Multilayer feed forward networks are
3.81 0.001 24.47 8 7.4 0.93 universal approximators. Neural Networks 2, 359e366.
4.09 0.0289 110.49 11 9.25 0.84 Ismael, N.F., Jeragh, A.M., 1986. Static cone tests and settlement of calcareous desert
10.55 0.039 8.65 15 13.2 0.88 sands. Canadian Geotechnical Journal 23 (3), 297e303.
6.79 0.0355 81.62 15 16.8 1.12 Jefferies, M.G., Davies, M.P., 1993. Use of CPTU to estimate equivalent SPT N60.
17.13 0.1005 43.18 21 25.2 1.20 Geotechnical Testing Journal, ASTM 16 (4), 458e468.
15.97 0.1468 40.96 34 31.3 0.92 Lunne, T., Robertson, P.K., Powell, J.J.M., 1997. Cone Penetration Testing in
7.89 0.02 25.3 12 14.1 1.18 Geotechnical Practice. Blackie Academic, EF Spon/Routledge Publishing, New
2.86 0.0146 71.08 6 7.0 1.17 York.
10.55 0.039 8.65 15 13.2 0.88 Masada, T., Sargand, S.M., Tarawneh, B., Mitchell, G.F., Gruver, D., 2006. New in-
spection and risk assessment methods for metal highway culverts in Ohio.
N60p: Predicted N60-value, N60f: Field N60-value. Transportation Research Records 1976 (1), 141e148.
204 B. Tarawneh / Geoscience Frontiers 8 (2017) 199e204
Masada, T., Sargand, S.M., Tarawneh, B., Mitchell, G.F., Gruver, D., 2007. Inspection
and risk assessment of concrete culverts under Ohio’s highways. Journal of
Performance of Constructed Facilities 21 (3), 225e233.
Meigh, A.C., Nixon, I.K., 1961. Comparison of in-situ tests of granular soils. In: Pro-
ceedings of 5th International Conference on Soil Mechanics and Foundation
Engineering, Paris, France.
Nejad, F.P., Jaksa, M.B., Kakhi, M., McCabe, B.A., 2009. Prediction of pile settlement
using artificial neural networks based on standard penetration test data.
Computers and Geotechnics 36 (7), 1125e1133.
Ni, S.H., Lu, P.C., Juang, C.H., 1996. A fuzzy neural network approach to evaluation of
slope failure potential. Journal of Microcomputers in Civil Engineering 11 (1),
59e66.
Principe, J., Euliano, N., Lefebvre, W.C., 1999. Neural and Adaptive Systems: Fun-
damentals through Simulations. Wiley, New York.
Robertson, P.K., 2009. Interpretation of cone penetration tests e a unified approach.
Canadian Geotechnical Journal 46 (11), 1337e1355.
Robertson, P.K., Campanella, R.G., Wightman, A., 1983. SPT-CPT correlations. Journal
of the Geotechnical Engineering Division 109 (11), 1449e1459.
Schmertmann, J.H., 1970. Static cone to compute static settlement over sand.
Journal of the Soil Mechanics and Foundations Division 96 (3), 1011e1043.
Shahin, M.A., 2014. State-of-the-art review of some artificial intelligence applica-
tions in pile foundations. Geoscience Frontiers 5 (6), 1e12.
Shahin, M.A., Jaksa, M.B., Maier, H.R., 2000. Predicting the Settlement of Shallow
Foundations on Cohesionless Soils Using Back-propagation Neural Networks.
Research Report No. R 167. The University of Adelaide, Adelaide, Australia.
Shahin, M.A., Maier, H.R., Jaksa, M.B., 2002. Predicting settlement of shallow
foundations using artificial neural networks. Journal of Geotechnical and Geo-
environmental Engineering 128 (9), 785e793.
Shahin, M.A., Jaksa, M.B., Maier, H.R., 2009. Recent advances and future challenges
for artificial neural systems in geotechnical engineering applications. Advances
in Artificial Neural Systems 2009, Article ID: 308239, 9 pages.
Shahri, A.A., Juhlin, C., Malemir, A., 2014. A reliable correlation of SPT-CPT data for
southwest of Sweden. Electronic Journal of Geotechnical Engineering 19,
1013e1032.
Shi, J., Ortigao, J.A.R., Bai, J., 1998. Modular neural networks for predicting settle-
ment during tunneling. Journal of Geotechnical and Geoenvironmental Engi-
neering 124 (5), 389e395.
Skempton, A.W., 1986. Standard penetration test procedures and the effects in
sands of overburden pressure, relative density, particle size, aging and over-
consolidation. Geotechnique 36 (3), 425e447.
Tarawneh, B., 2013. Pipe pile setup: database and prediction model using artificial
neural network. Soils and Foundations 53 (4), 607e615.
Tarawneh, B., 2014. Correlation of standard and cone penetration tests for sandy
and silty sand to sandy silt soil. Electronic Journal of Geotechnical Engineering
19, 6717e6727.
Tarawneh, B., Imam, R., 2014. Regression versus artificial neural networks: pre-
dicting pile setup from empirical data. KSCE Journal of Civil Engineering 18 (4),
1018e1027.
Tarawneh, B., Nazzal, M., 2014. Optimization of resilient modulus prediction from
FWD results using artificial neural network. Periodica Polytechnica 58 (2),
143e154.
Tatari, O., Sargand, S., Masada, T., Tarawneh, B., 2013. Neural network approach to
condition assessment of highway culverts: case study in Ohio. Journal of
Infrastructure Systems 19 (4), 409e414.
Zhou, Y., Wu, X., 1994. Use of neural networks in the analysis and interpretation of
site investigation data. Computer and Geotechnics 16 (2), 105e122.