Analysis of earthquake hazards prediction with multivariate adaptive regression splines

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 12, No. 3, June 2022, pp. 2885~2893
ISSN: 2088-8708, DOI: 10.11591/ijece.v12i3.pp2885-2893  2885
Journal homepage: http://ijece.iaescore.com
Analysis of earthquake hazards prediction with multivariate
adaptive regression splines
Dadang Priyanto1
, Muhammad Zarlis2
, Herman Mawengkang2
, Syahril Efendi2
1
Graduate Program of Computer Science, Department of Computer Science, Faculty of Computer Science and Information Technology,
Universitas Sumatera Utara, Medan, Indonesia
2
Department of Computer Science, Faculty of Computer Science and Information Technology, Universitas Sumatera Utara,
Medan, Indonesia
Article Info ABSTRACT
Article history:
Received May 1, 2021
Revised Jan 5, 2022
Accepted Jan 23, 2022
Earthquake research has not yielded promising results, either in the form of
causes or revealing the timing of their future events. Many methods have
been developed, one of which is related to data mining, such as the use of
hybrid neural networks, support vector regressor, fuzzy modeling, clustering,
and others. Earthquake research has uncertain parameters and to obtain
optimal results an appropriate method is needed. In general, several
predictive data mining methods are grouped into two categories, namely
parametric and non-parametric. This study uses a non-parametric method
with multivariate adaptive regression spline (MARS) and conic multivariate
adaptive regression spline (CMARS) as the backward stage of the MARS
algorithm. The results of this study after parameter testing and analysis
obtained a mathematical model with 16 basis functions (BF) and 12 basis
functions contributing to the model and 4 basis functions not contributing to
the model. Based on the level of variable contribution, it can be written that
the epicenter distance is 100 percent, the magnitude is 31.1 percent, the
location temperature is 5.5 percent, and the depth is 3.5 percent. It can be
concluded that the results of the prediction analysis of areas in Lombok with
the highest earthquake hazard level are Malaka, Genggelang, Pemenang,
Tanjung, Tegal Maja, Senggigi, Mangsit. Meninting, and Malimbu.
Keywords:
Conic multivariate
Multivariate
Non parametric
Peak ground acceleration
Prediction analysis
This is an open access article under the CC BY-SA license.
Corresponding Author:
Dadang Priyanto
Department of Computer Science, Faculty of Computer Science and Information Technology, Universitas
Sumatera Utara
Medan, Indonesia
Email: dadangpriyanto@students.usu.ac.id
1. INTRODUCTION
Earthquakes are natural disasters that can cause moderate to severe damage. Many lives and
property were lost as a result of the earthquake. Research on earthquakes to date has not provided significant
results to be able to determine the factors causing and or when the earthquake occurred. Many methods have
been developed in research related to earthquake prediction. In the field of computer science, research on
earthquake prediction is included in the scope of data mining research. In 2012, Han classifies the data
mining process into two groups, namely predictive data mining and descriptive data mining. Predictive data
mining is in principle a process of finding certain patterns and knowledge from big data sources [1]–[3]. A
mathematical function is needed in the data mining process, such as association, correlation, classification,
regression, and clustering functions [1]–[3]. Many methods are used in the predictive data mining process,
one of which is the multivariate adaptive regression spline (MARS) method [4], [5].

 ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 12, No. 3, June 2022: 2885-2893
2886
The MARS method is a non-parametric method that is very effective to overcome the problem of
high-dimensional data that is used to determine the relationship between predictor variables and response
variables. The problem in earthquake prediction is the existence of uncertain parameters and with the MARS
method the function of the mathematical model is influenced by the number of predictor variables used and
the maximum number of basis functions. Another factor is interactivity and minimum observations need to
be tested on the data used. The use of models in one area with other areas has different mathematical models
because in the analysis of earthquake predictions it is influenced by bedrock conditions, types of faults or
others. Predictive research with a nonparametric approach is preferred and has the advantage that this model
does not make specific assumptions regarding the underlying functional relationship between the responsive
variable and the predictor variable to estimate the general function of the high-dimensional data argument.
Prediction results are more effective even though the data set does not provide uniformity of information
from each earthquake recording station [6]–[8].
Previous research that uses MARS and conic multivariate adaptive regression splines (CMARS)
methods, such as the development of a robust computational method for data prediction problems with the
help of convex optimization (convex) in the presence of outliers in the dataset. The results show that the
optimal level of process parameters produces the desired response in the application. The research proposes a
new approach to deal with outliers in the prediction of ground motion in a systematic and effective manner.
The result is that there are no assumptions that must be validated for effective modeling in the presence of
outliers [9], [10].
Another study describes the development of a simple approach to predicting the displacement of
underground structures caused by earthquakes. The method used is the MARS model approach, to predict the
lift displacement of underground structures and evaluate the buoyancy of underground structures in terms of
earthquake parameters, structural characteristics, and soil properties [11]. Similar research on ground motion
prediction, explains that ground motion prediction equations (GMPEs) are empirical relationships used to
determine the response of the ground peak at a certain distance from the earthquake source. Research has
correlated the response of the ground peak as a function of the type of earthquake source, local conditions of
the location, distance from the source, depth, and magnitude of the earthquake strength. The method used is
CMARS on available datasets to obtain new GMPE. In the CMARS model, peak ground acceleration (PGA)
and peak ground velocity (PGV) values are used as dependent variables while three other parameters such as
magnitude moment (Mw), station location conditions (Vs30) and distance from earthquake source (Rjb) are
used as independent variables. This study shows that CMARS can be effectively used to predict PGA and
PGV values at various distances from the earthquake source [12], [13].
The main objective of this study is to analyze earthquake predictions using MARS and CMARS
involving 4 predictor variables with 16 maximum basis functions. This study will contribute to a
mathematical model of predictive analysis of earthquakes that occur in Lombok, West Nusa Tenggara,
Indonesia, which has different bedrock characteristics from other regions. This research using the MARS
method is the first research conducted in Lombok because earthquake prediction research at the same case
study location has never been done. The results of this study will classify areas that have a category prone to
earthquake hazards based on the highest PGA value.
2. RESEARCH METHOD
2.1. Multivariate adaptive regression spline (MARS)
The MARS method is a nonparametric regression method that is used to overcome the problem of
high-dimensional data, which is used to determine the pattern of the relationship between the response
variable and the predictor variable whose regression curve is not known and the previous information is not
complete enough [14]. Prediction data mining or called prediction analysis can be solved by two approaches,
namely parametric regression and nonparametric regression. These two approaches are commonly used as
statistical methods and are widely used as methods for investigating and modeling relationships between
variables [10]. The MARS method can overcome the shortcomings of recursive partitioning regression (RPR)
by producing a continuous model at knots and identifying the presence of an additive linear function. The
working system of the MARS method is a two-stage algorithm, namely the forward stepwise model and the
backward stepwise model [10], [15]. The first stage is the forward stepwise algorithm which is used to
combine the basis of function (BF), maximum interaction (MI), and minimum observation (MO) to find the
relationship between the respond variable and the predictor variable. Furthermore, the second stage of the
Backward Stepwise model is used as a simplification of the basis function (BF) obtained from the Forward
Stepwise stage. The basis function (BF) which has no contribution or makes a small contribution to the
Response variable will be eliminated at the backward stepwise model stage. This deletion process will have

Int J Elec & Comp Eng ISSN: 2088-8708 
Analysis of earthquake hazards prediction with multivariate adaptive regression splines (Dadang Priyanto)
2887
the effect of decreasing the number of squares of the least residual. In general, the nonparametric regression
model can be presented as in (1) [16]–[18]:
𝑦𝑖 = ƒ(𝑥𝑖) + ℰ𝑖 (1)
where 𝑦𝑖 = the dependent variable on observation 𝑖, ƒ(xi) = vector independent variable function, and ℰi =
is a free error 𝑖.
The determination of the independent variable greatly determines the results of the model built using
the MARS method so that the MARS model is flexible and its basic functions can be explained in (2) and (3):
(𝑥 − 𝑟)+ = {
𝑥 − 𝑟, 𝑖𝑓 𝑥 > 𝑟,
0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
(2)
and
(𝑥 − 𝑟)+ = {
𝑟 − 𝑥, 𝑖𝑓 𝑥 < 𝑟,
0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
(3)
As demonstrated (2) and (3) have almost the same function so they can be called the reflected pairs. The goal
is the reflected pairs in each variable xj for each observation xi,j on the knot of the variable. So that a
truncated linear function is formed from the base function as in (4):
𝑟 = {(𝑥𝑗 − 𝑟)+
, (𝑟 − 𝑥𝑗)+| 𝑟 ∈ {𝑥1𝑗, 𝑥2𝑗, … , 𝑥𝑁𝑗}, 𝑗 = 1,2, … , 𝑝} (4)
The MARS model starts from (5):
ƒ(𝑥) = 𝛽0 + ∑ 𝛽𝑚𝛽𝑚(𝑥),
𝑀
𝑚=1 (5)
where M is the number of basis functions that make up the model function 𝛽𝑚(𝑥) is the basis of the function
formed by a single element or by multiplying two or more elements contained in r, multiplied by the
coefficient 𝛽𝑚. The basic function to m can be explained into the base function as shown in (6):
𝛽𝑚(𝑥𝑚) = ∏ [𝑆𝑘𝑗
𝑚
(𝑋𝑘𝑗
𝑚
− 𝜏𝑘𝑗
𝑚
)]
𝐾𝑚
𝑗=1 +, (6)
where 𝐾𝑚 is the number of truncated linear functions times the base function to m. For 𝑋𝑘𝑗
𝑚
is the input
variable related to the truncated function in the base function to m. 𝜏𝑘𝑗
𝑚
is the knot variable value 𝑋𝑘𝑗
𝑚
.
Whereas 𝑆𝑘𝑗
𝑚
is the +/- operator which has a value of 1 or -1. The MARS model is flexible and is used to
overcome weaknesses in recursive partition regression by increasing the accuracy of the model. The MARS
model is run with two stages of the algorithm, namely forward stepwise and backward stepwise.
Furthermore, the algorithm will determine the knot value in a continuous model and drink the generalized
cross validation (GCV) value to obtain the best model. GCV measurement can be seen in (7):
𝐺𝐶𝑉(𝑀) =
1
𝑁
∑ [𝑦𝑖
𝑁
𝑖=1 −𝑓
̂𝑀(𝑥𝑖)]2
[1−
𝐶
̂(𝑀)
𝑁
]
2 (7)
where: 𝑦𝑖 = variabel dependent
𝑥𝑖 = variabel independent
N = the number of observations
𝑓
̂𝑀(𝑥𝑖) = the estimated value of the dependent variable on the M basis function on xi
M = maximum number of base functions
𝐶
̂(𝑀) = C(M)+d.M
𝐶(𝑀) = trace [B (BT
B)-1
BT
]+1; where B is a matrix of M basis functions
d = value when each base function reaches optimization (2 ≤ 𝑑 ≤ 4)
2.2. Conic multivariate adaptive regression splines (C-MARS)
C-MARS was developed as an alternative to the Backward Stepwise algorithm for the MARS
model. C-MARS with the letter "C" is a of "CONIC", convex or continuous, as shown in (8) [19]–[21]:

 ISSN: 2088-8708
2888
𝑃𝑅𝑆𝑆 = ∑ (𝑦𝑖 − 𝑓(𝑥𝑖))2
+ ∑ 𝜆𝑚 ∑ ∑ ∫ 𝛽
𝑚[𝐷𝑟,𝑠
𝛼 𝛽𝑚 (𝑡𝑚)]
2
𝑑𝑡𝑚
2
𝑟<𝑠
𝑟,𝑆∈𝑉𝑚
2
|𝛼|=1
𝛼=(𝛼1,𝛼2)𝑇
𝑀𝑚𝑎𝑥
𝑚=1
𝑁
𝑖=1
(8)
where (( 𝑥
̃𝑖 , 𝑦𝑖) 𝑖 = (1,2,3 … . 𝑁) represents the data points used as the predictor p-dimensional vector
variable 𝑥
̃𝑖 = (𝑥
̃𝑖 1, 𝑥
̃𝑖2 , … , 𝑥
̃𝑖𝑝)𝑇
(𝑖 = 1, 2, … , 𝑁) and response value N (𝑦𝑖, 𝑦2, … , 𝑦𝑁). Furthermore 𝑀𝑚𝑎𝑥
is the number of BF achieved at the end of the Forward MARS algorithm stage. Where: V(m)={(𝐾𝑗
𝑚
)|𝑗 =
1,2, … , 𝐾𝑚} is the set of variables associated with 𝑚𝑡ℎ
𝐵𝐹. 𝑧𝑚
= (𝑧𝑚1, 𝑧𝑚2, … , 𝑧𝑚𝑘𝑚
)𝑇
represents the
variables that contribute to 𝑚𝑡ℎ
𝐵𝐹 . 𝜆𝑚(𝑚 = 1,2, … , 𝑀𝑚𝑎𝑥) the value is always non-negative and is used as
a penalty parameter. 𝐷𝑟,𝑠
𝛼
𝛽𝑚(𝑡𝑚) ≔
𝜕|𝛼|𝛽𝑚
𝜕𝛼1 𝑡𝑟
𝑚 𝜕𝛼2 𝑡𝑠
𝑚 (𝑡𝑚), is the partial derivative for the basis function (BF)
to m. For 𝛼 = (𝛼1, 𝛼2)T
, | 𝛼| := 𝛼1 + 𝛼2 , where 𝛼1, 𝛼2 ∈ {0,1}
After simplifying the equation and adding the penalization of λ, for each derived term, using the
Tikhonov regularization the equation changes to be as (9):
𝑃𝑅𝑆𝑆 ≈ ‖𝑦 − 𝛽(𝑑
̃)𝜃‖2
2
+ 𝜆‖ 𝐿𝜃‖2
2
, (9)
Furthermore, it can be formulated into the conic quadratic problem (CQP) as shown in (10):
𝑚𝑖𝑛
𝑡, 𝜃
𝑡,
𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 ||𝛽(𝑑)𝜃 − 𝑦||2
≤ 𝑡,
||𝐿𝜃||2 ≤ √𝑀
̃
with t ≥ (10)
2.3. Peak ground acceleration (PGA)
PGA calculation is used to determine the maximum ground vibration acceleration that occurs in an
area caused by an earthquake. The PGA value can be obtained by empirical calculations using the attenuation
function of the Joyner and Boore attenuation equations [22], [23]. The first step is to determine the
coordinates of the location of a city in the area where the prediction analysis will be carried out, and the
vibration radius which is usually up to 500 km. The second step is to calculate the AVECOS value, which is
a value as a correction number because the Longitude coordinates towards the poles will be increasingly
different, and to calculate the AVECOS value with the formula as shown in formula (11) [23]:
𝐴𝑉𝐸𝐶𝑂𝑆 = 𝐶𝑜𝑠(𝑅𝐸𝐷𝐶𝑂𝑀 × 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝐿𝑎𝑡𝑖𝑡𝑢𝑑𝑒) (11)
Next, calculate the distance from the epicenter with the formula as in (12):
𝑅_𝑒𝑝𝑖 = √𝑥2 + 𝑦2 (12)
The epicenter is a seismic wave that is above the earth's surface and then spreads out in all directions. The
next step is to calculate using the Joyner and Boore attenuation function equations, as shown in (13):
𝑃𝐺𝐴 (𝑔𝑎𝑙) = 10(0,71+0,23(𝑀−6)−𝐿𝑜𝑔(𝑟)−0,0027.𝑟)
(13)
where M is the magnitude and 𝑟 is the root of (R_epi2
+82
)
𝑟 = √𝑅_𝑒𝑝𝑖2 + 82 (14)
by giving the value of 𝑀 and the value of 𝑟 in (13), the PGA value will be obtained.
2.4. Data set
The data used in this study were taken from the geophysics station Badan Meteorologi, Klimatologi,
and Geofisika (BMKG) of Mataram City. Data in the form of an earthquake catalog in Lombok with
positions (-4.0636
) south latitude (-13.0636
) south latitude and (111.5798
) east longitude (120.5798
) east
longitude for a period of 10 years between 2010 and 2019. Total earthquake data earth has a total of 8,053
records and varies in magnitude from 1.6-9.0 Mw, and a depth of 0-500 km [24], [25]. Magnitude data with a
value of less than 4.5 Mw and a depth of more than 300 km, is deleted because the data with this value does
not have a damaging impact.

2889
After selecting the data set based on magnitude 4.5 Mw and above and a depth of less than
300 kilometers, the results of data processing obtained a table of earthquake frequencies in Lombok with
grouping based on magnitude as shown in Table 1. The results of data processing can be seen in the graph of
the earthquake spread in Lombok based on the epicenter distance and magnitude strength as shown in
Figure 1.
Table 1. Earthquake frequency in Lombok is based on magnitude
No Magnitude (Mw) Frequency
1 4.5-5 283
2 5-6 121
3 6-7 15
Figure 1. Distribution of earthquakes in Lombok with magnitude 4.5-7
3. RESULTS AND DISCUSSION
3.1. PGA value calculation results
PGA is the maximum ground vibration acceleration that occurs in an area caused by an earthquake.
A large PGA value usually has a large impact or risk from an earthquake that occurs. The PGA value can be
obtained by one of them performing empirical calculations with the attenuation function. The attenuation
function is used to determine the relationship between the intensity of ground vibrations, the magnitude, and
the distance of an area from the source of the earthquake. There are several factors that affect the function of
attenuation, namely the earthquake mechanism, the distance of the epicenter and local soil conditions. This
research is to get the PGA value using the attenuation function of the Joyner and Boore attenuation equations
[22], [23]. From the results of data processing for earthquakes that occurred in Lombok from 2010 to 2019
the PGA values were obtained as shown in the example Table 2. After filtering and selecting the appropriate
variable, for the respond variable, namely PGA and the predictor variable, the depth (depth), magnitude
(Mw), epicenter distance (R-epi), and the addition of the variable temperature of the location of the incident
(SUHU), obtained ready data used for the predictive analysis process as shown in Table 3.
Table 2. PGA value for earthquake in Lombok
Time Long Lat Depth Mw AVECOS x y R-epi r log PGA PGA(g)
01-01-10 118.64 -9.5 99 3.8 0.987601365 281.1515923 -104.122929 299.812945 299.9196597 -3.082788016 0.000826441
01-01-10 119.06 -8.15 10 3.8 0.989382196 327.8645593 45.99022166 331.074417 331.1710581 -3.210214233 0.000616291
01-04-10 118.66-11.21 69 4.4 0.98514891 282.6442954 -294.266254 408.020129 408.0985493 -3.370631134 0.00042596
01-04-10 117.23 -8.29 29 4.5 0.989203895 126.5156216 30.42293193 130.122086 130.367777 -2.102163258 0.007903815
01-04-10 120.55 -7.32 527 5.1 0.990408927 492.2961883 138.2820108 511.348659 511.4112353 -3.5865806 0.000259071
01-05-10 119.07 -8.76 277 3.9 0.988594515 328.7028021 -21.8386836 329.427473 329.5245973 -3.180604251 0.000659775
01-07-10 119.07 -7.83 24 4.5 0.989784195 329.0983649 81.57259819 339.057256 339.1516218 -3.081103277 0.000829653
01-07-10 120.46 -7.94 23 3.9 0.989646878 482.0134675 69.34115626 486.975542 487.0412496 -3.775577119 0.000167657
01-09-10 118.43 -7.84 10 4.3 0.98977175 258.6573647 80.46064892 270.882905 271.0010117 -2.845673644 0.001426679
01-12-10 116.35 -8.75 155 3.4 0.988607654 29.70258763 -20.7267343 36.2193488 37.09233381 -1.557433461 0.027705535
13/01/2010 119.05 -9.83 17 4.2 0.987145193 326.025599 -140.817255 355.136862 355.2269563 -3.213618697 0.000611479
13/01/2010 117.1 -7.87 282 4.4 0.989734367 112.2765199 77.12480112 136.213993 136.4487151 -2.161380981 0.006896346
14/01/2010 119.83 -9.61 10 3.5 0.987450218 411.7699143 -116.354371 427.893447 427.9682255 -3.651925735 0.000222882
15/01/2010 118.52 -7.85 18 4.2 0.989759296 268.5591694 79.34869965 280.036147 280.1503946 -2.907797304 0.001236524

 ISSN: 2088-8708
2890
Table 3. Data respond variable and the predictor variable earthquake in Lombok
No Mw Depth SUHU (o
) R-epi PGA(g)
4 4.5 29 27.3 130.1220861 0.007903815
7 4.5 24 25.8 339.0572556 0.000829653
18 4.6 15 27.3 284.6781717 0.001460629
52 4.7 23 27 321.4371625 0.001085476
56 6.1 52 27.3 305.1784169 0.002654803
83 4.6 74 27.2 354.8631328 0.000757626
103 4.9 47 27.2 393.1406573 0.000631921
104 5 25 27.2 402.1835925 0.000615714
105 5 45 27.2 165.6369759 0.006495246
113 4.6 27 25.8 239.3908901 0.002301091
143 5.1 217 27 57.34264291 0.038371768
145 4.8 10 27 264.0445075 0.001990106
152 4.8 60 27 276.5372938 0.001758335
157 4.7 26 27 153.2842779 0.006463695
171 4.6 51 27 205.2507607 0.003317323
186 4.8 109 27 237.4462852 0.002610475
203 4.7 10 27.3 55.42425365 0.032480759
213 4.7 77 25.8 250.9637935 0.002153882
3.2. Results of the development of an analysis model for earthquake prediction in Lombok
Selection of the best MARS model after going through the forward stepwise algorithm and
backward stepwise algorithm based on a combination of BF, MI, and MO, the results of training data are
obtained. The results of MARS regression based on training data are as shown in Table 4. The results of
training data with a total of 16 basis functions BF after going through a process of elimination by applying
the Backward Stepwise algorithm, a total of 12 BF are formed, namely BF 1, 2, 3, 5, 7, 9, 10, 11, 13, 14, 15,
and 16. After doing elimination by eliminating the BF which does not have a contribution to the change in
the dependent variable, namely the BF 4, 6, 8, and 12 in order to obtain the best MARS model as in (15):
𝑌(𝑃𝐺𝐴) = −0.0175733 − 0.00211487 ∗ 𝐵𝐹1 + 0.0029936 ∗ 𝐵𝐹2 +
0.000556472 ∗ 𝐵𝐹3 + 0.00172513 ∗ 𝐵𝐹5 + 0.000373726 ∗ 𝐵𝐹7 +
0.000369563 ∗ 𝐵𝐹9 − 0.000160793 ∗ 𝐵𝐹10 − 0.000689482 ∗ 𝐵𝐹11 +
0.000676173 ∗ 𝐵𝐹13 + 0.00329239 ∗ 𝐵𝐹14 −
0.00125948 ∗ 𝐵𝐹15 + 6.46282𝑒 − 05 ∗ 𝐵𝐹16 (15)
Model PGA_G_=BF1, BF2, BF3, BF5, BF7, BF9, BF10, BF11, BF13, BF14, BF15, BF16
Table 4. Results of training data from MARS
W: 442.00 SQUARED: 0.99723
MEAN DEP VAR: 0.01402 ADJ R-SQUARED: 0.99715
UNCENTERED R-SQUARED=R-0 SQUARED: 0.99785
Parameter Estimate S.E. T-Value P-Value
Constant -0.01999 0.00723 -2.76347 0.00597
Basis Function 1 -0.00219 0.00022 -9.93648 0.00000
Basis Function 2 0.00307 0.00022 14.14851 0.00000
Basis Function 3 0.00056 0.00004 14.16816 0.00000
Basis Function 5 0.00180 0.00022 8.17230 0.00000
Basis Function 7 0.00037 0.00001 61.36239 0.00000
Basis Function 9 0.00037 0.0000 35.98398 0.00000
Basis Function 10 -0.00016 0.0000 -28.32933 0.00000
Basis Function 11 -0.00069 0.00009 -7.50372 0.00000
Basis Function 13 0.00067 0.00004 16.66312 0.00000
Basis Function 14 0.00326 0.00032 10.15419 0.00000
Basis Function 15 -0.00129 0.00020 -6.47462 0.00000
Basis Function 16 0.00006 0.00001 5.58802 0.00000
F-STATISTIC=12850.73516 S.E. OF REGRESSION=0.00138
P-VALUE=0.00000 RESIDUAL SUM OF SQUARES=0.00082
[MDF, NDF]=[12, 429] REGRESSION SUM OF SQUARES=0.29560
Based on the best MARS model obtained, independent variable inference that affects PGA based on
the MARS model according to the smallest GCV value in sequence based on the percentage of the
contribution is R-epi, Mw, SUHU, and depth. As shown in Table 5 which describes the interactivity of the

2891
contribution of independent variables to the dependent variable. It can be seen in Table 5 that the variables
that are very influential in the PGA value are the R-epi of 100% and the Mw of 31.08608%, while the SUHU
is 5.48525% and the depth amounting to 3.52988%.
Table 5. The interactivity of the independent variable contribution
Variable Importance GCV
R-EPI 100.00000 0.00067
MW 31.08608 0.00007
SUHU 5.48525 0.00000
DEPTH 3.52988 0.00000
3.3. Testing and validation
In prediction analysis required statistical analysis test to obtain hypothesis test results and determine
the level of significance. The significance level is intended to obtain the significance of the parameters [26].
Hypothesis testing is required to use statistical analysis to determine the significance of the parameters with
the suitability of the obtained mathematical model. This research is in testing the analysis of a mathematical
model using the partial regression coefficient test. In testing the partial regression coefficient, the following
formulation is required:
H0 : 𝑎1 = 𝑎2 = 𝑎3 = 𝑎5 = 𝑎7 = 𝑎8 = 𝑎9 = 𝑎11 = 0
H1 : there is at least one am ≠ 0;
m=1,2,3,5,7,9,10,11,13,14,15,16 (significant model)
- Significant level, =0.05
- Statistic test: tcount=
𝑎
̂𝑚
𝑆𝑒(𝑎
̂𝑚)
with 𝑆𝑒(𝑎
̂𝑚) = √𝑣𝑎𝑟 (𝑎
̂𝑚)
- Critical area: refuse H0 if 𝑡 > 𝑡(
𝛼
2
,61) or P-value <
P-value in statistical tests used to determine the magnitude of the opportunity, to state the status reject the
null hypothesis or (H0) with the actual condition (H0) is true.
As shown in Table 4 (results of training data) that the P-value is less than 0.05, or in other words,
every m< or (m<0.05) so that the H0 status is rejected. This means that each coefficient
𝛼1, 𝛼2, 𝛼3, 𝛼5, 𝛼7, 𝛼9, 𝛼10, 𝛼11, 𝛼13, 𝛼14, 𝛼15, 𝛼16 has a significant effect on the mathematical model obtained.
Based on the significance level of 5%, the mathematical model in formula (15) is significant, so it can be
used in predictive analysis of the PGA value for earthquake data sets in Lombok. Furthermore, after knowing
the suitability of the parameters and mathematical models obtained based on testing, it is concluded that the
variables that affect the PGA value are epicenter distance (R-epi), magnitude (Mw), temperature at the
location of the incident (SUHU), and depth (Depth).
3.4. Potential areas with the highest earthquake danger in Lombok
After going through the testing and validation of the results of the prediction analysis, it can be seen
that the areas in Lombok have the highest potential as shown in Table 6. The results of the calculation of the
PGA value will be influenced by the magnitude, depth, distance to the location of the incident, and the
temperature of the location where the earthquake occurred. In theory, based on a high PGA value, it will have
a high impact on earthquake damage, although there are other factors that influence earthquake damage such
as the condition of the bedrock of the location. Based on the results of predictive analysis by grouping areas
that have the highest earthquake hazard in Lombok, this data can be used by government policy makers to
make rules for infrastructure development with special specifications in earthquake-prone areas.
Table 6. Highest earthquake hazard potential areas in Lombok
No Time Lat Long Depth Mw R-epi PGA(g) PGV (cm/s) Temp (o
) Regional location
1 22-06-2013 -8.44 116.04 16 5.2 14.42381995 0.183715166 1.832770384 26.7 Malaka, Pemenang
2 09-08-2018 -8.36 116.22 12 6.2 27.39175594 0.16732396 1.814869245 24.9 Genggelang, Gangga
3 05-08-2018 -8.41 116.16 17 5.5 19.22254717 0.166068049 1.826111838 24.9 Tegal Maja, Tanjung
4 31/03/2016 -8.52 115.99 12 4.5 10.60647207 0.160605686 1.838096242 27.5 Senggigi, Meninting
5 06-08-2018 -8.42 116.03 23 5 16.8807379 0.143937973 1.829356187 24.9 Senggigi, Malimbu
6 04-05-15 -8.43 116.03 13 4.6 15.83302429 0.123357614 1.830810815 26.3 Mangsit, Senggigi

 ISSN: 2088-8708
2892
4. CONCLUSION
This research has obtained the result of a mathematical model involving a maximum number of
basis functions (BF) as many as 16 with 12 basis functions (BF) having a close contribution to the model.
The relationship between the predictor variable and the response variable has a contribution to the epicenter
distance of 100 percent, magnitude) of 31.1 percent, the temperature of the incident location by 5.5 percent
and the depth of 3.5 percent. Based on the highest PGA value, it can be concluded that the areas with the
highest level of earthquake hazard in Lombok are Pemenang, Malacca, Genggelang, Tegal Maja, Tanjung,
Senggigi, Mangsit. Meninting, and Malimbu. Earthquakes that have occurred in Lombok in the last 10 years
that have a high damage impact are in the category of shallow earthquakes because they have a depth of less
than 25 Kilometers.
ACKNOWLEDGMENT
Thank you very much to the Deputy for Strengthening Research and Development, Ministry of
Research and Technology/National Research and Innovation Agency (DRPM Kemenristek/BRIN) for Fiscal
Year 2020/2021. Thank you to the Chancellor of the University of North Sumatra, Dean and Head of
Doctoral Program in Computer Science, University of North Sumatra. Thank you to the Chancellor of
Bumigora Mataram University, who has provided full support for this research.
REFERENCES
[1] J. Han, M. Kamber, and J. Pei, Data mining: concepts and techniques. Morgan Kaufmann, 2012.
[2] T. Hastie, R. Tibshirani, and J. Friedman, “The elements of statistical learning,” New York, NY: Springer New York, 2009.
[3] D. T. Larose, Discovering knowledge in data: an introduction to data mining. John Wiley and Sons, 2005.
[4] F. Yerlikaya-Özkurt, İ. Batmaz, and G.-W. Weber, “A review and new contribution on conic multivariate adaptive regression
splines (CMARS): a powerful tool for predictive data mining,” in Springer Proceedings in Mathematics and Statistics, Springer
International Publishing, 2014, pp. 695–722.
[5] F. Yerlikaya, “A new contribution to nonlinear robust regression and classification with mars and its applications to data mining
for quality control in manufacturing,” Middle East Technical University, 2008.
[6] F. Yerlikaya-Özkurt, A. Askan, and G.-W. Weber, “An alternative approach to the ground motion prediction problem by a non-
parametric adaptive regression method,” Engineering Optimization, vol. 46, no. 12, pp. 1651–1668, Dec. 2014, doi:
10.1080/0305215X.2013.858141.
[7] K. M. Asim, A. Idris, T. Iqbal, and F. Martínez-Álvarez, “Earthquake prediction model using support vector regressor and hybrid
neural networks,” PLOS ONE, vol. 13, no. 7, Jul. 2018, doi: 10.1371/journal.pone.0199004.
[8] G. Asencio-Cortés, F. Martínez-Álvarez, A. Troncoso, and A. Morales-Esteban, “Medium–large earthquake magnitude prediction
in Tokyo with artificial neural networks,” Neural Computing and Applications, vol. 28, no. 5, pp. 1043–1055, May 2017, doi:
10.1007/s00521-015-2121-7.
[9] F. Yerlikaya-Özkurt, A. Askan, and G.-W. Weber, “A hybrid computational method based on convex optimization for outlier
problems: application to earthquake ground motion prediction,” Informatica, vol. 27, no. 4, pp. 893–910, Jan. 2016, doi:
10.15388/Informatica.2016.116.
[10] G.-W. Weber, İ. Batmaz, G. Köksal, P. Taylan, and F. Yerlikaya-Özkurt, “CMARS: a new contribution to nonparametric
regression with multivariate adaptive regression splines supported by continuous optimization,” Inverse Problems in Science and
Engineering, vol. 20, no. 3, pp. 371–400, Apr. 2012, doi: 10.1080/17415977.2011.624770.
[11] G. Zheng et al., “Evaluation of the earthquake induced uplift displacement of tunnels using multivariate adaptive regression
splines,” Computers and Geotechnics, vol. 113, Sep. 2019, doi: 10.1016/j.compgeo.2019.103099.
[12] R. A. Johnson and D. W. Wichern, Apllied Multivariate Statistical Analysis, 6th ed. Pearson, 2007.
[13] I. Peruš and P. Fajfar, “Ground-motion prediction by a non-parametric approach,” Earthquake Engineering & Structural
Dynamics, vol. 39, no. 12, 2010, doi: 10.1002/eqe.1007.
[14] R. L. Eubank, Nonparametric regression and spline smoothing. CRC Press, 1999.
[15] A. Özmen, “Robust conic quadratic programming applied to quality improvement a robustification of CMARS,” Middle East
Technical University, 2010.
[16] C. K. Arthur, V. A. Temeng, and Y. Y. Ziggah, “Multivariate adaptive regression splines (MARS) approach to blast-induced
ground vibration prediction,” International Journal of Mining, Reclamation and Environment, vol. 34, no. 3, pp. 198–222, Mar.
2020, doi: 10.1080/17480930.2019.1577940.
[17] C. Yazıcı, F. Yerlikaya-Özkurt, and İ. Batmaz, “A computational approach to nonparametric regression: bootstrapping CMARS
method,” Machine Learning, vol. 101, no. 1–3, pp. 211–230, Oct. 2015, doi: 10.1007/s10994-015-5502-3.
[18] J. H. Friedman, “Multivariate adaptive regression splines,” The Annals of Statistics, vol. 19, no. 1, Mar. 1991, doi:
10.1214/aos/1176347963.
[19] A. Özmen, İ. Batmaz, and G.-W. Weber, “Precipitation modeling by polyhedral RCMARS and comparison with MARS and
CMARS,” Environmental Modeling and Assessment, vol. 19, no. 5, pp. 425–435, Oct. 2014, doi: 10.1007/s10666-014-9404-8.
[20] A. Çevik, G.-W. Weber, B. M. Eyüboğlu, and K. K. Oğuz, “Voxel-MARS: a method for early detection of Alzheimer’s disease
by classification of structural brain MRI,” Annals of Operations Research, vol. 258, no. 1, pp. 31–57, Nov. 2017, doi:
10.1007/s10479-017-2405-7.
[21] J. F. H. Jr., W. C. Black, B. J. Babin, and R. E. Anderson, Multivariate data analysis, 7th ed., vol. 50, no. 181. Prentice Hall
International, 2010.
[22] W. B. Joyner and D. M. Boore, “Prediction of earthquake response spectra.” US Geological Survey, pp. 359–375, 1982, doi:
10.3133/ofr82977.
[23] W. Joyner and D. M. Boore, “Measurement, characterization and prediction of strong ground motion,” in Earthquake Engineering

2893
and Soil Dynamics II - Recent Advances in Ground-Motion Evaluation: Proceedings of the Specialty Conference, 1988,
pp. 43–102.
[24] M. N. Shodiq, D. H. Kusuma, M. G. Rifqi, A. R. Barakbah, and T. Harsono, “Neural network for earthquake prediction based on
automatic clustering in Indonesia,” JOIV : International Journal on Informatics Visualization, vol. 2, no. 1, Feb. 2018, doi:
10.30630/joiv.2.1.106.
[25] “Indonesia tsunami early warning system,” BMKG, 2019. https://inatews.bmkg.go.id/new/view_event_eq.php?catid=5953
(accessed Nov. 16, 2019).
[26] D. R. Cox et al., “The role of significance tests,” Scandinavian Journal of Statistics, vol. 4, no. 2, pp. 49–70, 1977.
BIOGRAPHIES OF AUTHORS
Dadang Priyanto Degree (S3) Doctoral Program in Computer Science at the
Universitas Sumatera Utara in 2021, Master Degree (S2) in Computer Science obtained at
Universitas Gadjah Mada Yogyakarta in 2006, and Bachelor Degree (S1) in Informatics
Engineering obtained at STMIK Akakom Yogyakarta in 1998. Currently works as a permanent
lecturer at Bumigora University Mataram, in the Software Engineering study program.
Research has been carried out since 2005 in the fields of Multimedia and Data Science. The
latest publication is related to Data Mining for Earthquake Predictions in Lombok. He can be
contacted by email: dadang.priyanto@universitasbumigora.ac.id.
Muhammad Zarlis is a Professor of Computer Science at the Universitas
Sumatera Utara, A Doctoral Degree (S3) in Computer Science was obtained at the University
Sains Malaysia, Malaysia and a Masters Degree (S2) in Computer Science was obtained at the
University of Indonesia-Sandwich Program with the University of Maryland USA, and a
Bachelor Degree (S1) in Physics was obtained at the University of North Sumatra. Research
that has been pursued for a long time is related to the fields of Intelligent Systems, Computer
Security, and Computation. He can be contacted by email: m.zarlis@usu.ac.id.
Herman Mawengkang is a Professor of Mathematics at the Universitas Sumatera
Utara. Masters and Doctoral degrees were obtained at the University of New South Wales
Australia, and Bachelor degrees (S1) in Mathematics were obtained at the Universitas
Sumatera Utara. The research that has been done so far is related to the fields of Mathematics
and Operations Research. He can be contacted by email: hmawengkang@yahoo.com.
Syahril Efendi is a Permanent Lecturer at the Computer Science Doctoral
Program at the Universitas Sumatera Utara. A Doctorate Degree (S3) in Mathematics was
obtained at the Universitas Sumatera Utara, a Masters Degree (S2) in Computer Science was
obtained at the National University of Malaysia, Malaysia, and a Bachelor Degree (S1) in
Mathematics was obtained at the Universitas Sumatera Utara. The research he has been
pursuing to date is related to the fields of research operations and information science. He can
be contacted by email: syahril1@usu.ac.id.

Analysis of earthquake hazards prediction with multivariate adaptive regression splines

More Related Content

Similar to Analysis of earthquake hazards prediction with multivariate adaptive regression splines (20)

More from IJECEIAES (20)

Recently uploaded (20)

Analysis of earthquake hazards prediction with multivariate adaptive regression splines