Journal of Computer Science

Original Research Paper

A New Algorithm for Earthquake Prediction Using Machine

Learning Methods
1
Nada Badr Jarah, 2Abbas Hanon Hassin Alasadi and 3Kadhim Mahdi Hashim
1Department of Computer Science, College of Computer Sciences and Maths, University of Kufa, Iraq
2Department of Computer Information Systems, College of Computer Sciences and Information Technology,
University of Basrah, Basrah, Iraq
3Department of Technology Engineering College of Information Technology, Imam Ja'afar Al-Sadiq University, Baghdad, Iraq

Article history Abstract: Seismic tremors are among the foremost perilous normal fiascos
Received: 07-07-2023 individuals confront due to their event without earlier caution and their
Revised: 09-09-2023 effect on their lives and properties. In expansion, to consider future disaster
Accepted: 19-10-2023 prevention measures for major earthquakes, it is necessary to predict
Corresponding Author:
earthquakes using Neural Networks (NN). A machine learning technique
Nada Badr Jarah has developed a technology to predict earthquakes from ground controller
Department of Computer data by measuring ground vibration and transmitting data by a sensor
Science, College of Computer network. Devices to process this data and record it in a catalog of seismic
Sciences and Maths, University data from 1900-2019 for Iraq and neighboring regions, then divide this data
of Kufa, Iraq into 80% training data and 20% test data. It gave better results than other
Email: [email protected] prediction algorithms, where the NN model performs better Seismic
prediction than other machine learning methods.

Keywords: Earthquakes, Neural Networks, Machine Learning, Prediction,

Earthquakes Data

Introduction researchers in the field of earthquake prediction in

different parts of the world.
Seismic tremor is a fundamental normal wonder Sathwik et al. (2022) employed a set of machine
influencing organisms' life and their property. It is the learning algorithms to predict future earthquakes such as
sudden release of energy transmitted by waves from the logistic regression, Support Vector Machine (SVM),
ground. It destroys vast areas in a few minutes. It leads to
random forest classifier, and k-nearest neighbors. The
considerable losses of lives and property and predicting
examined dataset consists of 14 features to protect the
earthquakes gives at least a little time to protect people
assets of the residencies and the best efficiency reached
and reduce earthquake damage (Bilal et al., 2022).
over this study was around 0.9. This result was very high
This research is considered the first informational
because of the number of training features in comparison
study to predict earthquakes in Iraq, as the recording of
with the predicted details.
earthquake data readings after installing seismic
Bangar et al. (2020) mixed the random forest and
monitoring stations since 1914 in Iraqi territory. In support vector machine algorithms to detect early signs of
addition to data recorded by international devices since the earthquake, the tested dataset was related to India with
1900, many earthquakes in Iraq and areas far from the the rest of the neighboring countries and all data were
seismic fault line raised Geologists' fear of foreseeing an from government sources such as the united states
increment in seismic tremors for the coming long time geological survey and the India meteorological
(Jarah et al., 2023; Rouet‐Leduc et al., 2017). department. The accuracy of this study was enclosed to
This research applied a technique to predict earthquakes 0.74, 0.76, and 0.83 to provide the government with
using the data from several seismic monitoring stations. important details about future earthquakes.
Machine learning calculations were utilized to prepare and Mallouhy et al. (2020) utilize different eight machine
analyze a design of information to anticipate the event of a learning algorithms which are Random Forest (RF), Naive
seismic tremor and powerful computational techniques Bayes (NB), Logistic Regression (LR), Multi-Layer
emerged from analyzing big data. Perceptron (MLP), AdaBoost (AB), Support Vector
This study is the first in Iraq to use local data collected Machine (SVM) and Regression Trees (RT). The
in Iraqi catalogs. The following are studies conducted by examined dataset was public data related to world

© 2024 Nada Badr Jarah, Abbas Hanon Hassin Alasadi and Kadhim Mahdi Hashim. This open-access article is distributed
under a Creative Commons Attribution (CC-BY) 4.0 license.
Nada Badr Jarah et al. / Journal of Computer Science 2024, 20 (2): 150.156
DOI: 10.3844/jcssp.2024.150.156

earthquakes and consisted of 139 features. Unfortunately,

LR recorded the worst results over these algorithms which
identified 52 wrong results over 139. The best accurate
prediction values were archived by the SVM algorithm
around 0.7 to provide the authority parties with the
required information about future earthquakes.
Asencio-Cortés et al. (2018) combined several
regression algorithms with ensemble learning
algorithms to analyze the big data to identify the
earthquake magnitude in the coming seven days in
California only. This machine learning was used to
analyze 1 Gigabyte dataset gathered over the duration
between 1970-2017 by using R language and Amazon
cloud infrastructure and the highest accuracy recorded
over the examination rounds is 0.8.
Asim et al. (2018) used four machine learning
techniques to predict the magnitude of upcoming
earthquakes in Hindukush which are pattern recognition
neural network, recurrent neural network, random forest, Fig. 1: Locations of earthquake monitoring stations
and linear programming boost ensemble classifier. The
Earthquakes Data
dataset consists of 441 vectors, each vector related to one
month of the duration from 1977-2013. The prediction Earthquakes are recorded with a device installed in
results were around 0.58, 0.64, 0.62and 0,65, the best an underground vault and at nine stations connected to a
accuracy meted by LPBoost ensemble technology. wireless sensor network whose center is the receiver at
We compared the proposed model with the results of the Seismological Laboratory of the University of
others above, where the proposed model obtained 0.83 as Basrah (SLUB). Figure 1 shows the locations of seismic
a verification result while the others were separated at this stations. And the network address of these stations
upper and lower value. Mallouhy et al. (2020) recorded according to their location on the Internet page at the
0.76 as a distinct result, while Asim et al. (2018) reported source (Fernandes et al., 2022).
0.64. One of the featured studies was published higher The seismic recording device is called a broadband. As
than the proposed model, with the same other researchers it records all waves or any human activity that leads to
(2022) scoring 0.9, while Bangar et al. (2020) achieved earthquakes that reach it, regardless of distance and
0.83, which is what was recorded in the study. Recently magnitude, since 2014, it has converted earthquakes into
(2022) Southwick's results were found to be better than electrical signals recorded and analyzed by the computer
(Liu et al., 2018).
the proposed model for several reasons such as the
Broadband operates on the principle of inertia, which
number of features for the training model is 14 incoming
involves sensing motion relative to the movement of the
vectors with the proposed model only having 3 directions
ground and is recorded by a digitizer.
for the training model and this has more significant
One of the most significant challenges is that most of
implications for a distinct model. Bangar et al. (2020)
the databases created by the geological centers need to
carried out the study and checked the model on an Indian adjust the timings and other details so that we can use
database and the number of training options was 6 vectors, them for the field of machine learning.
arriving at a result of 0.83. However, the model was tested
in 5 directions and the size of the current dataset consists Machine Learning
of 34.663 records which is less than the size of the dataset There are multiple algorithms in machine learning,
used by Bangar et al. (2020). Another independent each with its strengths and weaknesses, so when solving
research on top of the proposed model produced by a problem, first consider the necessary algorithm.
Asencio-Cortés et al. (2018) recorded strength even in the If it is still challenging to decide, the decision closest
next seven days in California, the main reason being to the problem-solving process can be selected from the
related to the large data set used in this research which actual use example (Probst and Boulesteix, 2017).
produced 600 million data records between 1970 and The following are the three machine learning
1970. and 2019, showing comparable trend trends with algorithms that were closest to predicting earthquakes that
the 34.663 records in our sample. passed the earthquake data in this study.

151
Nada Badr Jarah et al. / Journal of Computer Science 2024, 20 (2): 150.156
DOI: 10.3844/jcssp.2024.150.156

Table 1: Application of LR, RFR, and NN algorithms

No. Input features Predicted features LR RFR NN
1 Timestamp, latitude, magnitude Depth, longitude 0.182 0.401 0.675
2 Timestamp, depth, longitude Magnitude, latitude 0.249 0.658 0.839
3 Timestamp, latitude, depth Magnitude, longitude 0.199 0.585 0.685
4 Timestamp, latitude, longitude Magnitude, depth 0.250 0.398 0.489
5 Depth, longitude, magnitude Timestamp, latitude 0.218 0.666 0.654
6 Latitude, depth, longitude Magnitude, timestamp 0.122 0.441 0.702
7 Latitude, depth, magnitude Timestamp, longitude 0.252 0.577 0.758
8 Timestamp, longitude, magnitude Depth, latitude 0.116 0.477 0.634
9 Longitude, latitude, magnitude Depth, timestamp 0.180 0.301 0.768
10 Timestamp, depth, magnitude Longitude, latitude 0.114 0.495 0.775

Table 2: The catalog of Earthquake in Iraq and surrounding regions (1900-2019) (Onur et al., 2017)
Eventide Year Month Day Time Lat Lon Depth (km) MAG Timestamp
10001 1900 2 24 06:00.0 38.45 44.8700 0.0 5.4 -2204312256
10002 1900 4 17 17:00.0 38.00 46.0000 0.0 6.2 -2200442256
10004 1901 2 6 48:00.0 33.00 49.0000 0.0 7.4 -2174266656
10005 1901 5 20 38:36.0 38.38 42.2300 10.0 5.5 -2165401296
| | | | | | | | |
| | | | | | | | |
44659 2019 12 29 28:27.4 38.8709 43.5207 14.3 2.1 -1577582824
44660 2019 12 30 56:21.8 30.8205 50.0474 10.0 3.7 -1577682728
44662 2019 12 31 50:57.3 32.4992 46.9521 10.0 2.8 -1577750223
44663 2019 12 31 49:12.8 34.7548 45.5659 12.2 2.8 -1577743923

Linear Regression covered-up layers between the two layers of the input
and the outside layer (Alves, 2006). A NN that contains
A linear regression model was applied to the Iraqi data more than one hidden layer is known as a profound
set. The result shown in Table 1. shows that the linear neural arrangement and learning is called profound
regression was not significant and could not meet any learning. Each layer has a boundless number of hubs.
useful prediction. A test showed the worst prediction, with The number of hubs within the input layer breaks even
verification values ranging from 0.114-0.252. Thus, linear with the number of input information highlights.
regression cannot predict the earthquake based on the Covered-up layers comprise a boundless number of
current data and the parameters used. Although this model hubs. The yield layer comprises as it were one hub for
was not achieved, the required prediction must use other understanding relapse issues and more than one hub for
techniques to find the best prediction (Ranjan et al., 2019). evaluating pictures, translation, audio, or solving
Random Forest Regressor classification issues (Tapia-Hernández et al., 2019). In
Table 1. Application of the above three algorithms with
It is a mechanism that generates multiple decision tree different probabilities in input and output features.
models and decides on the final prediction by the majority
vote of those models. Because each model trains using Earthquake Prediction
sub-datasets generated by splitting the original training
Seismic tremors are a portion of earth's life and an
dataset, each model exhibits slightly different prediction
appalling portion of human history and they are sudden
performance. Therefore, the more it is generated, the more
Decision tree models, the generalization ability of the unsettling influences within the earth's outside, as not a
random group as a whole increases, and the prediction year goes by without hearing handfuls of seismic tremors,
performance better (González et al., 2019). a few of them solid and destroying, wiping out whole
cities. Consequently, the issue of anticipating their event
Neural Network appears exceptionally imperative.
NN may be a machine learning strategy or Earthquake forecasting is the science of determining
calculation that attempts to reenact the working of the details of impending earthquakes in terms of the
neurons within the human brain for learning. At first, the location, magnitude, and time of the earthquake within a
results appear inaccurate and after a specific iteration of given area (Mavrodiev et al., 2018).
the data, they adjust themselves so that the results Researchers face challenges in predicting earthquakes
increase in accuracy. Moreover, each NN is organized and dealing with natural phenomena. According to the
within the frame of layers of counterfeit cells: An inward study of (Li et al., 2022), there's no valid forecast within
layer, an external layer, and layers between them, or the brief term since the reason for the short-term

152
Nada Badr Jarah et al. / Journal of Computer Science 2024, 20 (2): 150.156
DOI: 10.3844/jcssp.2024.150.156

expectation is to empower emergency measures to The data is divided into determining the input X and
diminish activity and pulverization, which leads to untrue featured three as input (latitude, longitude, and time
desires and dissatisfaction within the occasion of a critical stamp). Output Y has the features that predict them as
seismic tremor, resulting in additional losses and legal output (Magnitude and Depth), where the best-fit
penalties (Kiani et al., 2019).
parameters of the show are utilized to calculate the result
Measurable speculation testing strategies and utilizing preparing and test information.
performing machine learning approaches, to be specific
Finally, the experimental practical step applies the
polynomial calculated relapse and bolster Vector Machine
machine learning methods. The data of 44540 earthquakes
(SVM) for seismic tremor information, may be utilized in
is divided into a training set of 80% and a testing set of
classification and relapse investigation to decide the
20%. Thus, it is 33405 for training and 11135 for testing.
likelihood of a seismic tremor (Onur et al., 2017). Three types of algorithms are learned as follows.
Proposed Algorithm In the beginning, two types of linear regression are
used: The first method of ML, which starts with Linear
The first step of the proposed algorithm reads the
Regression to start predicting the data in the first
earthquake data file in Iraq and the surrounding area from
experiment of the model and according to the instruction reg,
1900-2019, As in Table 2.
fit and it contains the Train and y-train and the expectation is
Furthermore, the second step is data preprocessing,
test. It means that the random forest regress or model or
which represents the six main features in earthquake data:
machine learning method is not successful.
Earthquake date (year, month, and day), time, longitude,
Secondly, using the second ML method, which is
latitude, depth, and earthquake strength.
random forest regress, and applying the training process,
The date and time data in this mechanism are not
testing on test and Y_testand when implementing. The
considered digital data. They cannot be addressed in a
credibility of 0.6 is fitted, meaning that the model fails.
completed program owing to the complexity of the
Furthermore, the expected values are not correct.
format, the reading process, and the training process with
Finally, the third method, NN, includes model building
the existence of (:) In time. So, it is changed to digital data
and model execution, i.e., building the model first, and
only, i.e., to Unix time, which is in seconds, where it can
after preparing it, implementing it. The model is in two
be easily used as an input to the network we have created.
types: The main one always used is called sequential and
By merging the date and time data, converting its
the dense is the starting point of the model, so the model
FORMAT, storing it in a variable called time stamp, and
type is sequential and adding the first layer, which is
adding it to the database, they all were negative because
dense and needs activation and loss function and Input
they represent the past time.
shape = 3, so it has three entries for this model.
Thus, the data will be dealt with after processing to be:
Thus, building an NN model, according to the above
Time stamp, longitude, latitude, depth, and earthquake
method, has several features, namely: Loss function,
strength. Part of it can be illustrated in Table 3.
optimizer, activation, and metrics that represent four
The locations of earthquakes in the study area are
variables that must be implemented to build this model
indicated on the map, as shown in Fig. 2 (Wathiq et al., 2020).
here we have a process based on this data, the model
A fundamental problem appeared, which is the missing
should give the best options because there are several
data. Each implementation gives a significant error rate and
options, that is, Building the model requires several
the model cannot be trained. Therefore, a mechanism was
factors, namely:
adopted to delete all the missing data whose value is null to
filter the earthquakes in Iraq. We used them in the training
process and 123 rows were deleted. a) The batch size can be either 10, 20, 50, or 100
The third step signifies the application of the b) Activation may be one of the following options:
algorithm; when the database is ready to form step two, 'Relu', 'tanh', 'sigmoid', 'hard sigmoid', 'linear ', and
data has been read and arranged to determine the fields 'exponential'
that will be worked upon in the data Table 3. It has c) Optimizer, in which the result is checked for each
become limited to only five variables. cell. If it is weak, it returns to the input, adjusts
The third step signifies the application of the the weights, and repeats the process of checking
algorithm; when the database is ready to form step two, the output. It has several types: ('Stochastic
data has been read and arranged to determine the fields Gradient Descent (SGD)', 'RMSprop', 'Adagrad',
that will be worked upon in the data Table 3. It has 'Adadelta', ' Adam', 'Adamax', 'And Nadam') and
become limited to only five variables. each has several options

153
Nada Badr Jarah et al. / Journal of Computer Science 2024, 20 (2): 150.156
DOI: 10.3844/jcssp.2024.150.156

Thus, an experiment will be made to choose the best

in the model by trying all the options and then it will print
the best options that this model can achieve based on this
database, which is as follows.
Activation is relu, the optimizer is SGD, and the loss
function is squared_hinge it takes accuracy in metrics.
A suitable NN has been built for the training data set
using Unsupervised learning, using 16 nodes after testing
with four. Then eight, the results were not satisfactory and in
the case of 16, the results were acceptable. Figure 3 shows
the components of the NN used to predict earthquakes.
The NN consists of the following layers: 3-layer Input,
16-layer hidden, 16-layer hidden, and 2-layer output. Fig. 3: NN in earthquake prediction

Fig. 4: Machine learning algorithms

It is executed 20 times for training on the data and we

can increase it so that it does not go beyond overfitting.
The accuracy value was equal to 0.5 and when the epochs
were increased and became, for example, 40 times. We
started the training process again. The accuracy started to
equal 0.5. We continued Increasingly, it reached an
accuracy: Of 0.839 and the execution of the program
stopped and we got a loss function value equal to 0. 003
Fig. 2: Locations of earthquakes in the study area on the map in Fig. 4. Performance of the proposed models.

Table 3: Part of the essential data for earthquakes in Iraq from 1900-2019

Latitude Longitude DEPTH (km) Magnitude Timestamp

38.4500 44.8700 0.0 5.4 -2204312256

38.0000 46.0000 0.0 6.2 -2200442256
33.0000 49.0000 0.0 7.4 -2174266656
38.3800 42.2300 10.0 5.5 -2165401296
| | | |
| | | |
38.8709 43.5207 14.3 2.1 -1577582824
30.8205 50.0474 10.0 3.7 -1577682728
32.4992 46.9521 10.0 2.8 -1577750223
34.7548 45.5659 12.2 2.8 -1577743923

154
Nada Badr Jarah et al. / Journal of Computer Science 2024, 20 (2): 150.156
DOI: 10.3844/jcssp.2024.150.156

Materials and Methods Acknowledgment

To evaluate the performance of the proposed model Thank you to the publisher for their support in the
with regard to forecasting efficiency, the materials publication of this research article. We are grateful for the
consisted of running the proposed scheme on a data set resources and platform provided by the publisher, which
of 119 years with a size of (34663×14). Our have enabled us to share our findings with a wider
experiments were conducted on an Intel i7-1065G7 audience. We appreciate the efforts of the editorial team
processor at 2.5 GHz; with windows 11 64-bit, 16GB in reviewing and editing our work, and we are thankful for
RAM (HP laptop). Furthermore, we used Python the opportunity to contribute to the field of research
Colabe to conduct our experiments. through this publication.
As for the research methods, it can be represented in
Fig. 5 to illustrate the building of the model: Funding Information
The authors have not received any financial support or
funding to report.

Author’s Contributions
Nada Badr Jarah: Collect data and written the research.
Abbas Hanon Hassin Alasadi: The idea of research,
designed and analysis of data.
Kadhim Mahdi Hashim: Written the program and
analyzing the results.

Ethics
Fig. 5: Earthquake prediction model building
This article is original and contains unpublished
Results and Discussion material. The corresponding author confirms that all of the
other authors have read and approved the manuscript and
The results of applying the three machine learning no ethical issues involved.
algorithms, namely LR, RFR, and NN, to earthquake data
in Iraq and the surrounding areas are as shown in Table 1.
References
The five features of earthquakes used in this study were
switched, and Fig. 4 shows the accuracy in using the three Alves, E. I. (2006). Earthquake forecasting using neural
algorithms to predict earthquakes. networks: Results and future work. Nonlinear
Dynamics, 44, 341-349.
Conclusion https://doi.org/10.1007/s11071-006-2018-1
Asencio-Cortés, G., Morales-Esteban, A., Shang, X., &
In recent years, Iraq has witnessed continuous
activity of earthquakes in the governorates of Iraq. This Martínez-Álvarez, F. (2018). Earthquake prediction
study is considered a major challenge and its data are in California using regression algorithms and cloud-
limited, as we need a larger number of inputs to obtain based big data infrastructure. Computers and
an accurate prediction of the outputs. In this study, we Geosciences, 115, 198-210.
find that the number of inputs is 3 and the number of https://doi.org/10.1016/j.cageo.2017.10.011
outputs is 2. The highest number of inputs was Asim, K. M., Idris, A., Iqbal, T., & Martínez-Álvarez, F.
achieved. The accuracy result of the machine learning (2018). Earthquake prediction model using support
models, which is the neural network, reached about vector regress or hybrid neural networks. PloS
0.839 using the seismic data catalog in Iraq and the One, 13(7), e0199004.
surrounding areas for 119 years (1900-2019). The https://doi.org/10.1371/journal.pone.0199004
inputs are Timestamp, Depth, and Longitude to predict Bangar, P., Gupta, D., Gaikwad, S., Marekar, B., &
Magnitude and Latitude. The proposed model was Pati, J. (2020). Earthquake prediction using
compared with models proposed by others, we found machine learning algorithm. International Journal
that the proposed model achieves appropriate accuracy of Recent Technology and Engineering
depending on the number of training features available (IJRTE), 8(6), 4684-4688.
as well as the size of the data set. https://doi.org/10.35940/ijrte.E9110.018620

155
Nada Badr Jarah et al. / Journal of Computer Science 2024, 20 (2): 150.156
DOI: 10.3844/jcssp.2024.150.156

Bilal, M. A., Ji, Y., Wang, Y., Akhter, M. P., & Yaqub, Mavrodiev, S. C., Pekevski, L., Botev, E., Pinar, A.,
M. (2022). An early warning system for earthquake Kikuashvili, G., & Gilat, A. (2018). Study of the
prediction from seismic data using Batch Normalized Possibility of Predicting Earthquakes. International
Graph Convolutional Neural Network with Attention Journal of Geosciences, 9(12), 688-706.
Mechanism (BNGCNNATT). Sensors, 22(17), 6482. https://doi.org/10.4236/ijg.2018.912042
https://doi.org/10.3390/s22176482 Onur, T., Gök, R., Abdulnaby, W., Mahdi, H., Numan, N.
Fernandes, M., Corchado, J. M., & Marreiros, G. (2022). M., Al‐Shukri, H., ... & Abd, N. A. (2017). A
Machine learning techniques applied to mechanical comprehensive earthquake catalog for Iraq in terms
fault diagnosis and fault prognosis in the context of of moment magnitude. Seismological Research
real industrial manufacturing use-cases: A systematic Letters, 88(3), 798-811.
literature review. Applied Intelligence, 52(12), https://doi.org/10.1785/0220160078
14246-14280. Probst, P., & Boulesteix, A. L. (2017). To tune or not to tune
https://doi.org/10.1007/s10489-022-03344-3 the number of trees in random forest. The Journal of
González, J., Yu, W., & Telesca, L. (2019). Earthquake Machine Learning Research, 18(1), 6673-6690.
magnitude prediction using recurrent neural https://www.jmlr.org/papers/volume18/17-269/17-269.pdf
networks. Multidisciplinary Digital Publishing Ranjan, G. S. K., Verma, A. K., & Radhika, S. (2019). K-
Institute Proceedings, 24(1), 22. nearest neighbors and grid search cv based real time
https://doi.org/10.3390/IECG2019-06213 fault monitoring system for industries. In 2019 IEEE
Jarah, N. B., Hashim, K. M., & Alasadi, A. H. (2023). 5th International Conference for Convergence in
Earthquake prediction in Iraq using machine learning Technology (I2CT) (pp. 1-5). IEEE.
techniques Indonesian. Journal of Electrical https://doi.org/10.1109/I2CT45611.2019.9033691
Engineering and Computer Science, vol. 29, no. 1, Rouet‐Leduc, B., Hulbert, C., Lubbers, N., Barros, K.,
pp. 322-329. Humphreys, C. J., & Johnson, P. A. (2017).
https://doi.org/10.11591/ijeecs.v29.i1.pp322-329 Machine learning predicts laboratory
Kiani, J., Camp, C., & Pezeshk, S. (2019). On the earthquakes. Geophysical Research Letters,
application of machine learning techniques to derive 44(18), 9276-9282.
seismic fragility curves. Computers and Structures, https://doi.org/10.1002/2017GL074677
218, 108-122. Sathwik, G., Patil, V. S. S. Br, Vishesh, S. Yatish, J.
https://doi.org/10.1016/j.compstruc.2019.03.004 Soham S. (2022). Machine Learning Approach for
Li, L., Kang, Y., Petropoulos, F., & Li, F. (2022). Feature-
Predicting Earthquakes in a Geographic Location.
based intermittent demand forecast combinations:
International Journal of Advanced Research in
Accuracy and inventory implications. International
Journal of Production Research, 1-16. Computer and Communication Engineering
https://doi.org/10.1016/j.ijforecast.2021.11.001 (IJARCCE), vol. 11, no. 11, pp. 110-115.
Liu, H., Liu, Z., Liu, S., Liu, Y., Bin, J., Shi, F., & Dong, https://doi.org/10.17148/IJARCCE.2022.111122
H. (2018). A nonlinear regression application via Tapia-Hernández, E., Reddy, E. A., & Oros-Avilés, L.
machine learning techniques for geomagnetic data J. (2019). Earthquake predictions and scientific
reconstruction processing. IEEE Transactions on forecast: Dangers and opportunities for a technical
Geoscience and Remote Sensing, 57(1), 128-140. and anthropological perspective. Earth Sciences
https://doi.org/10.1109/TGRS.2018.2852632 Research Journal, 23(4), 309-315.
Mallouhy, R., Abou Jaoude, C., Guyeux, C., & Makhoul, https://doi.org/10.15446/esrj.v23n4.77206
A. (2020). Major earthquake event prediction using
Wathiq, A., Tuna, O., Rengin, G., Shakir, A. M., Hanan,
various machine learning algorithms. In 2019
International Conference on Information and M., Al-Shukri, H., ... & Ali, R. (2020). Probabilistic
Communication Technologies for Disaster seismic hazard assessment for Iraq. Journal of
Management (ICT-DM) (pp. 1-7). IEEE. Seismology, 24(3), 595-611.
https://doi.org/10.1109/ICTDM47966.2019.9032983 https://doi.org/10.1007/s10950-020-09919-2

156