Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/341635348

LSTM-based Models for Earthquake Prediction

Conference Paper · March 2020


DOI: 10.1145/3386723.3387865

CITATIONS READS
28 751

3 authors, including:

Asmae Berhich Mohammed Issam Kabbaj


Mohammadia School of Engineers Mohammadia School of Engineers
7 PUBLICATIONS 51 CITATIONS 33 PUBLICATIONS 372 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Asmae Berhich on 21 August 2022.

The user has requested enhancement of the downloaded file.


LSTM-based Models for Earthquake Prediction
Asmae BERHICH, Fatima-Zahra BELOUADHA, Mohammed Issam KABBAJ
AMIPS research team, E3S research center
Mohammadia school of engineers, Mohammed V University in Rabat, Morocco
berhich.asmae@gmail.com, belouadha@emi.ac.ma, kabbaj@emi.ac.ma

ABSTRACT
Over the last few years, many works have been done in earthquake The application of ANN, RNN, and DNN models is present in
prediction using different techniques and precursors in order to earthquake prediction where many works have emerged using
warn of earthquake damages and save human lives. Plenty of works different techniques, models and data sets of different areas.
have failed to sufficiently predict earthquakes, because of the However, many of them were not capable to make a reliable
complexity and the unpredictable nature of this task. Therefore, in prediction especially in the case of large earthquakes. They
this work we use the powerful deep learning technique. A useful couldn’t capture the correlations in the datasets because of their
algorithm that captures complex relationships in time series data. small number in the studied areas.
The technique is called long short-term memory (LSTM). The work In this paper, we present our work of earthquake prediction using
employs this method in two cases of study; the first learns all the the data sets of Morocco since it is not immunized from the risks of
datasets in one model, the second case learns the correlations on earthquakes and their disastrous consequences. Earthquakes
two divided groups considering their range of magnitude. The generate significant human and material damage and their costs are
results show that learning decomposed datasets gives more well- in billions of Dirhams. The case of Al Hoceima, which experienced
functioning predictions since it exploits the nature of each type of a tragedy in 2004 is an example. The data set of Moroccan
seismic events. seismicity from 1900 to 2019 is given by the National Geophysical
Institute of the National Centre for Scientific and Technical
CCS Concepts Research CNRST.
• Computing methodologies → Machine learning → Machine
learning approaches → Neural networks. This work is based on the application of the deep learning technique
Long short-term memory (LSTM), which is widely used to classify,
Keywords process and predict time series data problems. LSTM is a variant of
Prediction; earthquakes; LSTM; time series data; deep learning. recurrent neural networks known by their ability to model sequence
data and to remember past data in memory.
1. INTRODUCTION
Earthquakes can suddenly strike any region in the world; they lead By this powerful algorithm, we model our seismic datasets in two
to great damages depending on their magnitudes. Earthquakes with cases of study. The first one gives the prediction of the magnitude,
large magnitudes could be potentially fatal and cause serious location and year of the incoming earthquakes using the whole
economic and material losses. The medium earthquakes are also dataset, the second case focuses on datasets decomposition based
dangerous especially for countries that are not taking the necessary on their magnitude range. The decomposition makes two groups of
precautions. Warning from earthquakes was and is still a the dataset for prediction. The first group presents large
challenging problem that needs more in-depth researches and earthquakes, and the second presents medium and small
suitable solutions. Many works on earthquake predictions have earthquakes. The models of both cases are evaluated and compared
been done for many years. Even that it was considered an in the following sections. Section 2 presents the classification of
impossible task, machine learning and deep learning challenges previous works that have been applied in earthquake prediction.
have allowed realizing this task and making it possible. From 1994 The LSTM model architecture is explained in section 3. Section 4
till now seismologists and scientists are trying to solve the describes our datasets and introduces our proposed methodology
earthquake prediction using neural networks and deep learning and model. The performance of our models is evaluated and
[19]. However, a lot of works do not fill the full meaning of discussed in section 5. Finally, the last section concludes in brief
earthquake prediction. Seismologists had defined earthquake the aim of our paper.
prediction by providing [1]:
2. RELATED WORKS
a specific time range, a specific location, a specific magnitude In this section, we present the previous works classified into four
range and the probability that performs the prediction. categories:

Permission to make digital or hard copies of all or part of this work for Works based on precursor signals, this technique uses the natural
personal or classroom use is granted without fee provided that copies are effects and their abnormal behaviors, for example, Fan et al [6]
not made or distributed for profit or commercial advantage and that copies present an earthquake prediction approach based on extracting the
bear this notice and the full citation on the first page. Copyrights for texture and emergence frequency of clouds and estimate the
components of this work owned by others than ACM must be honored. possible location of earthquakes. Florido et al [7] analyze large
Abstracting with credit is permitted. To copy otherwise, or republish, to seismic events of Chilean zones to detect precursory patterns for
post on servers or to redistribute to lists, requires prior specific permission
large earthquakes using clustering algorithms. Hayakawa [9]
and/or a fee. Request permissions from Permissions@acm.org.
NISS2020, March 31-April 2, 2020, Marrakech, Morocco Propose to utilize the electromagnetic phenomena as a short-term
© 2020 Association for Computing Machinery. precursor and presented the history and the reason for using these
ACM ISBN 978-1-4503-7634-1/20/03...$15.00 precursors. Hayakawa and Yamauchi et al [10] monitor the milk
https://doi.org/10.1145/3386723.3387865
yield of cows at Kagawa, and they find an abnormal depletion about to find the best architecture of the magnitude earthquake prediction
10 days before an earthquake, they discuss and compare this model. Narayanakumar and Raja [21] use seismic indicators and
behavior with electromagnetic precursors. historical data, and evaluate the performance of Back-propagation
(BP) neural networks to predict earthquakes in the region of
Works based on statistical and mathematical approaches like in Himalaya. The proposed model is a three-layer feed-forward BP
[14] Kannan identify patterns within the random occurrences of ANN. Finally, Panakkat and Adeli [24] propose an RNN model for
multiple seismic zones like California, central USA, Northeast earthquake location and time prediction of moderate to large
USA, Hawaii, Turkey, and Japan using Poisson distribution and earthquakes using seismic, considering two cases of studies
spatial connection model. Pasari [25] employs probability location decomposition and time decomposition.
distributions such as Weibull and gamma models to observe the
cumulative probability of magnitude 6.0 or higher in the northern We notice that most of the abovementioned neural network models
Himalaya. Sitharam et al [26] use three models lognormal, Weibull use various kinds of features as input to predict the time and
and gamma to estimate the probability of occurrences in six magnitudes of earthquakes, but none of them considers the
different seismic regions of different tectonic features. They find decomposition of magnitude ranges, and the correlations are not
that a higher value of ln L guarantees a performant model using the well studied either.
logarithmic probability of the likelihood function. Boucouvalas et
al [5] produce a modified version of the mathematical technique 3. MODEL ARCHITECTURE
FDL based on Fibonacci, Dual and Lucas numbers, in order to 3.1. Recurrent neural networks architecture
predict the location of the epicenter of an earthquake. Recurrent neural networks (RNNs) are a class of artificial neural
works based on machine learning and artificial neural networks networks (ANNs) specified by their memory state.
(ANN), such as Su and Zhu [29] who build an ANN model to study The RNNs network is similar to ANN's one, where the model
the correlations between the maximum of earthquakes affecting calculates the output by multiplying inputs with weights and the
coefficient and influencing factors like basemen rock and site activation function, in order to add the non-linearity to the network
condition. Ni and Zhou [22] apply A model for damage detection, (see Figure 2). In contrast, RNNs considers the memorized output
where they applied a principal component analysis (PCA) based of the previous time step t-1 and add it to the inputs of current time
data reduction technique to the measured frequency response step t, and this is the role of their memory state (equation 1 and 2).
functions FRFs as the input variables of ANN instead of the raw
FRF data. Asim et al [2] elaborate a machine learning classifiers- ht=tanh (Whhht-1+Wxhxt) (1)
based model, it utilizes the eight calculated features from the yt=Whyht (2)
geophysical parameters of Gutenberg-Richter’s inverse law. Four
classifiers are applied and compared neural networks, recurrent where Whh is the weight of the previous hidden state h t-1, xt is the
neural networks, random forest, and linear programming boost current input, Wxh is the weight of the current input state and tanh
ensemble classifiers in order to predict earthquakes magnitude for is the activation function. yt is the output state and Why is the weight
the Hindukush. Moreover, Asim et al [3] Compute seismic at the output state.
indicators to consider the maximum of information of seismic
activity in different regions, then they apply the genetic The memory is the key of RNNs which allows them to learn the
programming and Adaboost (GP-Adaboost) as an ensemble correlations in sequence data, where it examines the whole context
method to predict earthquakes of magnitude 5.0 and above. And in and elements of each timestep to make predictions.
[20] the same authors calculate sixty seismic features using RNNs are applicable in time series data because they are dependent
geophysical and seismological concepts, then a support vector on each other, which present behavior and the trend change by time
machine (SVM) regressor combined by a hybrid neural network in their sequence values.
(merging three different ANNs) are employed to predict
earthquakes in three different regions. Furthermore, it exists some But, RNNs become sometimes untrainable since they suffer from
ANN approaches applied to precursors signals to predict the vanishing and exploding gradient problem. When the
earthquakes. For instance, Moustra et al [20] predict the magnitude information is passing in long timesteps and deep layers, the
of the impending seismic events using artificial neural networks gradients can't progress and converge then the model cannot learn
and the seismic electric signals as features for input data, because and the gradient stays constant.
they are believed to occur before an earthquake and considered as
earthquake precursors. Itai et al [13] introduce a multilayer neural 3.2. Long short-term memory architecture
network using compression data to detect precursor signals from LSTM is an RNN that replaces the standard neural network layer
the electromagnetic waves. These waves radiate from the earth’s with LSTM cells, proposed by Hochreiter [11] in 1991.
crust and they are useful for earthquake prediction. Külahci et al The LSTM cells are enhanced by three components called gates:
[16] build A three-layer Levenberg-Marquardt feedforward the input gate, the forget gate and the output gate, its architecture is
learning algorithm using eight different parameters including radon illustrated in Figure 2.
gas changes for earthquake prediction. Ozerdem et al [23]
Elaborate a neural network model to extract correlations between LSTM trains the features in a different fashion where it starts by
Spatio-temporal electric field data in order to detect hazard using the tanh activation function to squash the input data and make
precursory anomalous signal patterns. them very small in a very non-linear manner. After that, the features
are passed into the input gate which takes the relevant information
Works based on deep learning i.e. Li and Liu [17] suggest a from the squashed input by multiplying them with a sigmoid
combination of backpropagation neural networks and an improved function, this function filters the elements that are not required
variant of particle swarm optimization for magnitude prediction, where the values are between 0 (remove from the network) and 1
for improving PSO they used a non-linear decreasing inertia weight (pass through the network). Afterward, another important element
strategy. Mahmoudi et al [18] develop 128 different MLP networks called the internal state is the memory of the current state. It takes
into account the previous state s t-1 and adds it to the input data (as descriptive statistics of our data, where the largest magnitude is 7.3
in RNN). It uses an addition operation instead of multiplication to and the smallest one is 0.02.
avoid the vanishing problem. These operations are described in the
following equations: 4.2. Methodology
Earthquakes events are a time series data, that are not captured by
The recurrence of states is enforced by a forget gate, this gate linear and classical methods, but with LSTM complex architectures
decides which state elements should be memorized or forgotten can be successfully trained and predicted with multiple input
using a sigmoid function. variables. Therefore, our research is based on this model in order to
Finally, the Tanh function squashes the outputs. These outputs are find an efficient and performant result in earthquake prediction.
controlled by an output gate that specifies the values that are In this section, we present the important steps of our work to predict
allowed to be the outputs of the current cell state. earthquakes in Moroccan regions.
it= σ (Wi . [ht-1, xt] + bi) (3) Initially, we start by introducing the first case of our work, which
uses the model LSTM and the whole dataset we have presented
ct=tanh (Wc [ht-1, xt] + bc) (4) above.
ft= σ (Wf . [ht-1, xt ] + bf) (5) In the second case, we apply the LSTM model on decomposed data
ot= σ ( Wo [ht-1, xt] + bo) (6) based on their magnitude range. In the two cases of studies, we
predict the magnitude, location and year of the incoming
ht=ot * tanh (ct) (7) earthquakes.
where it, ct, ft, ot, ht are the input gate, cell state, forget gate, output The Flow chart in Figure 3 presents our proposed LSTM model to
gate, and the hidden state respectively. Wi, Wc, Wf and Wo are their predict earthquakes. In the first step, the datasets are normalized
weight matrices respectively. bi, bc, bf, and bo are the biases. Xt is using the Min-Max scaler, this scaler is a transformation technique
the input, ht-1 is the last hidden state, ht is the internal state. σ is the calculated by formula 8, it transforms datasets to an exact same
sigmoid function. scale, in a range between 0 and 1. Such transformers are used to
standardize the features so that no one dominates the others. The
4. LSTM MODEL FOR EARTHQUAKE second step consists of dividing data into 80% of the training set
PREDICTION and 20% of the testing set. Then the model is trained and supported
In this section, we present the dataset studied in our work, the by the mean squared error (MSE) for error calculation and
evaluation, and Adam optimizer for convergence to the minimum
preprocessing used techniques and finally we explain our
methodology and describe our LSTM models architecture. error. Adam is widely used in deep learning because it helps the
model to achieve good results fast. In [15] empirical research
4.1. Dataset demonstrates that Adam works well and it outperforms other
In this work, we use the dataset of seismic activity of the regions of stochastic optimizers. Finally, when the model is trained, the
Morocco recorded from 1900 to 2019. The data was gathered from predictions are computed on the testing set, and the gap between
the national geophysical institute of CNRST. predicted values and real values is calculated using the Mean
squared error and the Mean absolute error (MAE). The MAE and
4.1.1. Data preprocessing MSE, are used to evaluate regression models and were used in
After cleaning, removing duplicated data and merging data files previous works of earthquakes prediction. Their calculations are
into one file, we set 10 features which present: described in formula 9 and 10 respectively.
• The geographic characteristics Latitude and 𝑥−min⁡(𝑥)
z = max(𝑥)−min(𝑥) (8)
Longitude of the seismic event
• The depth of the event by kilometers 𝑁
1
• Day, Month, Year, Hour, Minute and Second: we keep 𝑀𝐴𝐸 = 𝑁 ∑ | 𝑦𝑖 − 𝑦̂| (9)
1
all the elements of the time when the events occurred, to 𝑖=1
conserve the exact information. 𝑁
• Mag is the magnitude of the seismic event 1
𝑀𝑆𝐸 = 𝑁 ∑ (𝑦𝑖 − 𝑦̂)2 (10)
1
𝑖=1
Some negative magnitudes presented as outliers in our dataset are
removed since they don’t cause any damage and they are not felt. 4.2.1. Study case 1: Earthquake prediction using
Feature generation or other possible added features are not treated,
LSTM
since the ANN and deep learning models especially LSTM are In this case, we are using the 29689 seismic events, that we present
capable to extract and learn original and complex data without in 4.1.2, where we apply our proposed architecture of the LSTM
using any other tool to generate their characteristics. For instance, algorithm illustrated in figure 4. The proposed architectures are
authors in [4] compared the situations when using the features with found after doing a tuning approach to search the most adequate
generated characteristics like b-value and using original features, architecture. The way used to tune is suggested in [27], where they
thereby they found the same results which demonstrate the recommend finding the balance between underfitting and
capability of deep learning to learn insights by itself. overfitting with an examination of training and testing loss.
As it is described in figure 4, our proposed architecture contains a
4.1.2. Data description LSTM layer, a dropout function, a dense network, and reLu
After data preparation, we get 29689 from the 32396 seismic events activation function. The LSTM layer contains 15 neurons which
in which we will apply our model directly. Table 1 presents the give the optimal result after trying multiple numbers for this case.
The dropout function is applied to LSTM layer. It is used to drop
out of the network some neurons, where it doesn't consider them earthquake, the MSE is 0.11 and MAE is 0.027, for small
during the training process. This function helps the model to avoid earthquakes, the MAE is 0.041 and the MSE is 0.0058 and the
overfitting since it ignores the co-dependency between neurons overall error are calculated where the MAE is 0.042 and MSE is
[28]. This limits the power of each neuron to deal with its 0.0059 (Table 2). Training time in the second case is faster since it
calculation for new inputs individually, and focus on historical ends in 6.27 seconds before the first case’s model. Hence, the
data. decomposition of datasets by magnitudes gives more performant
The dense network is a fully connected neural network, that is results, because the lack of datasets for large earthquakes makes
connected to the output neurons of the LSTM layer. And it is used them hard to learn. But, the power of LSTM in learning
to give the desired targets predicted from the output of the pattern complicated data realizes the extraction of patterns from the few
of the LSTM layer. The dense network applies an activation datasets of large earthquakes, especially when learning them
function for outputs calculation. In our case, we use the Rectified individually. The fitting curves of our models are illustrated in
Linear Function (reLu) as it is simple and performant. It returns the Figure 5, all the models are well fitted and not overfitted since we
main value if it is positive and 0 if it is null or negative. This use the dropout function.
function is recommended by Goodfellow and Bengio [12] where Comparing with last works is very hard in the field of earthquake
they say: ‘’ major algorithmic change that has greatly improved the prediction because of the use of different performance metrics and
performance of feedforward networks was the replacement of different dataset, catalogs, and regions [8]. Consequently, to
sigmoid hidden units with piecewise linear hidden units, such as evaluate the performance of our LSTM models we build ANN
rectified linear units.”. The outputs of the model are presented in models with same architectures for both cases, Table 3 shows the
four features the Magnitude, Longitude, Latitude and Year of the results of each case. We choose the algorithm ANN since it is
coming earthquakes. widely used in literature as it is presented in Section 2. The LSTM
models are outperforming the ANN models in both cases.
4.2.2. Study case 2: Earthquake prediction using Especially when predicting large earthquakes, ANN gives 0.30 for
LSTM and magnitude decomposition MAE and 0.13 for MSE. ANN doesn’t show any difference in
In this case, we consider two LSTM-based models applied on performance when decomposing datasets since it wasn't able to
divided datasets. The decomposition of our dataset is based on their well predict large earthquakes as LSTM did. Whereas, the training
range of magnitudes since the characteristics of large earthquakes time for decomposed data is slow by 37.62 seconds.
couldn’t be related to the smallest ones. There is a big gap between In brief, the results of our experiments prove the performance and
the two seismic magnitude ranges. Large, medium and small effectiveness of our LSTM models in earthquakes prediction. In
earthquakes do not present the same problem; each one have his addition, our models are not complicated by using seismic
own type of dangers, patterns, and typical features, e.g. we could indicators and generated features. The LSTM can learn patterns and
never study rich and poor people shopping activities in one model features from datasets by itself. Furthermore, our work fills the full
because we think it is the same case of our seismic datasets. meaning of earthquakes prediction where it gives all the above-
Hence, we decompose the datasets into two different magnitude mentioned important elements in section 1, the magnitude, location
ranges, as follows: and time. Then, we recommend to use and evaluate our models in
earthquake prediction with similar datasets and seismic activity.
• Small and medium earthquakes: From magnitudes 0.2 to
under 5.0; 6. CONCLUSION
• Large earthquakes: magnitude 5.0 and above.
In this paper, we have suggested a new model for earthquake
The flowcharts in Figure 4 present our used architectures for each prediction using historical datasets. We have built two model
model, which we found after testing multiple different ones. The prediction architectures. The first is an LSTM model that we apply
LSTM layer contains 15 neurons for small and medium on the dataset of Moroccan regions. It predicts year, location and
earthquakes, and 10 neurons for large earthquakes. The activation magnitude of earthquakes. The second one focuses on datasets
function applied in both models is reLu and a dropout function is decomposition based on their range of magnitudes and applies two
applied in both models. LSTM models on the divided data. The decomposition we propose
The outputs of the models are presented in four features the is performant and efficient especially when predicting large
Magnitude, Longitude, Latitude and Year of the coming earthquakes. Experiment results of our LSTM models are described
earthquakes. and evaluated and compared with ANN models. The results
demonstrate that our proposed model achieves favorable
5. PERFORMANCE EVALUATION AND performance compared to others.
DISCUSSION
In this section, we evaluate the performance of each model in both
cases of studies using the MAE and MSE metrics. First, we start by
evaluating the performance of earthquake prediction when using
the whole dataset in one model. After that, we evaluate the
performance of the prediction when decomposing the datasets into
two different magnitude ranges. Finally, we discuss and compare
the results of the two cases, and we evaluate their performance
against ANN models with the same architectures.
Figure 1. Typical architecture of Recurrent neural networks
The simulation results of our model in the first case give good algorithm
results where the MAE is 0.075 and the MSE is 0.014 as shown in
table 3. But, in the second case when data is decomposed into two
groups results become much better and improved; for large
Figure 2. Typical architecture of Long-short term memory
algorithm

Figure 4. Flow charts of LSTM models used in earthquake


prediction. The model in the left is applicated when using all
datasets, and the other two models are applicated when using
decomposed datasets.

Figure 3. Flow chart of the proposed LSTM model

Figure 5. Plotting results of prediction models with and without


datasets decomposition

Table 1. Descriptive statistics of seismic dataset of Morocco from 1900 to 2019 after data preparation
Characte
Depth Latitude Longitude Year Month Day Hour Minute Second Mag
ristics
count 29689 29689 29689 29689 29689 29689 29689 29689 29689 29689
24.15938 34.61972 2000.95 11.4719
mean -5.384475 5.884166 15.900805 29.421368 29.106403 2.410389
1 3 4865 93
20.5470 6.93243
std 25.23118 1.777117 25.008664 3.505600 9.048841 17.397553 17.608949 0.879518
24 9
min 0.100000 20.02000 -4251.00000 1901.00 1.000000 1.000000 0.00000 0.000000 0.000000 0.020000
25% 10.00000 33.52400 -6.550000 1991.00 3.000000 8.000000 5.00000 14.000000 14.000000 1.700000
18.07185
50% 35.25000 -4.100000 2009.00 6.000000 16.000000 12.0000 29.000000 29.000000 2.400000
3
75% 30.00000 35.66600 -3.676000 2016.00 9.000000 24.000000 17.0000 44.000000 44.000000 2.940540
max 675.0000 43.32000 7.648000 2019.00 12.000000 31.000000 23.0000 59.000000 59.000000 7.300000
infrared cloud images. (Dec. 2015), 98150E.
Table 2. Experiment results of LSTM models, when using all [7] Florido, E. et al. 2015. Detecting precursory patterns to
datasets and when using decomposed magnitude ranges using enhance earthquake prediction in Chile. Computers and
the performance metrics MAE and MSE, and the elapsed time Geosciences. 76, (Mar. 2015), 112–120.
during training process. DOI:https://doi.org/10.1016/j.cageo.2014.12.002.
[8] Galkina, A. and Grafeeva, N. 2019. Machine learning
Total for the methods for earthquake prediction: A survey. CEUR
ALL Range Range
Metrics decomposed Workshop Proceedings. 2372, (2019), 25–32.
Dataset [0.2, 0.5[ [0.5,0.2[
datasets
[9] Hayakawa, M. 2016. Earthquake prediction with
0.075612 0.041774 0.11138
MAE 0.042091154 electromagnetic phenomena. AIP Conference
26 616 759
Proceedings (Feb. 2016).
0.014995 0.005821 0.02728
MSE 0.005918795 [10] Hayakawa, M. et al. 2016. On the Precursory Abnormal
616 1996 5077
86.63059 15.6931 Animal Behavior and Electromagnetic Effects for the
Training 61.66931 77.362433300 Kobe Earthquake (M~6) on April 12, 2013. Open Journal
59224700 1920000
time 41 00002 of Earthquake Research. 05, 03 (2016), 165–171.
9 0012
DOI:https://doi.org/10.4236/ojer.2016.53013.
Table 3. Experiment results of ANN models, when using all [11] Hochreiter, J. 1991. DIPLOMARBEIT IM FACH
datasets and when using decomposed magnitude ranges using INFORMATIK Untersuchungen zu dynamischen
the performance metrics MAE and MSE, and the elapsed time neuronalen Netzen.
during training process. [12] Ian Goodfellow, Yoshua Bengio, A.C. 2017. The Deep
Learning Book. MIT Press. 521, 7553 (2017), 785.
Total for the DOI:https://doi.org/10.1016/B978-0-12-391420-0.09987-
ALL Range Range
Metrics decomposed X.
Dataset [0.2, 0.5[ [0.5,0.2[
datasets
0.072536 0.081498 0.30554 [13] Itai, A. et al. 2005. Multi-layer neural network for
0.0825543178 precursor signal detection in electromagnetic wave
MAE 81569956 21554646 2783127
852341 observation applied to great earthquake prediction. (Sep.
802 755 6546
0.012023 0.015362 0.13679 2005), 31–31.
0.0159352242
MSE 64744513 80415918 7647679 [14] Kannan, S. 2014. Innovative Mathematical Model for
96990257
1583 4586 3874 Earthquake Prediction. Engineering Failure Analysis. 41,
75.60912 5.79379 (2014), 890–895.
Training 43.78028 81.402928000
97000000 8300000 DOI:https://doi.org/10.1016/j.engfailanal.2013.10.016.
time 88 00002
1 006 [15] Kingma, D.P. and Ba, J.L. 2015. Adam: A method for
stochastic optimization. 3rd International Conference on
Learning Representations, ICLR 2015 - Conference Track
7. REFERENCES Proceedings (2015).
[1] Allen, C.R. 1976. Responsibilities in earthquake
prediction. Bulletin of the Seismological Society of [16] Külahcı, F. et al. 2009. Artificial neural network model for
America. 66, 6 (1976), 2069–2074. earthquake prediction with radon monitoring. Applied
Radiation and Isotopes. 67, 1 (Jan. 2009), 212–219.
[2] Asim, K.M. et al. 2017. Earthquake magnitude prediction
DOI:https://doi.org/10.1016/J.APRADISO.2008.08.003.
in Hindukush region using machine learning techniques.
Natural Hazards. 85, 1 (Jan. 2017), 471–486. [17] Li, C. and Liu, X. 2016. An improved PSO-BP neural
DOI:https://doi.org/10.1007/s11069-016-2579-3. network and its application to earthquake prediction.
Proceedings of the 28th Chinese Control and Decision
[3] Asim, K.M. et al. 2018. Seismic indicators based
Conference, CCDC 2016 (Aug. 2016), 3434–3438.
earthquake predictor system using Genetic Programming
and AdaBoost classification. Soil Dynamics and [18] Mahmoudi, J. et al. 2016. Predicting the Earthquake
Earthquake Engineering. 111, (Aug. 2018), 1–7. Magnitude Using the Multilayer Perceptron Neural
DOI:https://doi.org/10.1016/j.soildyn.2018.04.020. Network with Two Hidden Layers. Civil Engineering
Journal. 2, 1 (Jan. 2016), 1–12.
[4] Bhatia, A. et al. 2018. Earthquake forecasting using
DOI:https://doi.org/10.28991/cej-2016-00000008.
artificial neural networks. International Archives of the
Photogrammetry, Remote Sensing and Spatial [19] Mignan, A. and Broccardo, M. 2019. Neural Network
Information Sciences - ISPRS Archives. 42, 5 (2018), 823– Applications in Earthquake Prediction (1994-2019):
827. DOI:https://doi.org/10.5194/isprs-archives-XLII-5- Meta-Analytic Insight on their Limitations. September
823-2018. (2019), 1–25.
[5] Boucouvalas, A.C. et al. 2015. Modified-Fibonacci-Dual- [20] Moustra, M. et al. 2011. Artificial neural networks for
Lucas method for earthquake prediction. Third earthquake prediction using time series magnitude data or
International Conference on Remote Sensing and Seismic Electric Signals. Expert Systems with
Geoinformation of the Environment (RSCy2015) (Jun. Applications. 38, 12 (Nov. 2011), 15032–15039.
2015), 95351A. DOI:https://doi.org/10.1016/j.eswa.2011.05.043.
[6] Fan, J. et al. 2015. Research on earthquake prediction from [21] Narayanakumar, S. and Raja, K. 2016. A BP Artificial
Neural Network Model for Earthquake Magnitude adjoining regions. Geomatics, Natural Hazards and Risk.
Prediction in Himalayas, India. Circuits and Systems. 07, 9, 1 (Jan. 2018), 568–588.
11 (2016), 3456–3468. DOI:https://doi.org/10.1080/19475705.2018.1466730.
DOI:https://doi.org/10.4236/cs.2016.711294. [26] Sitharam, A.S.T.G. and Haider, S.T. 2015. Probabilistic
[22] Ni, Y.Q. et al. 2006. Experimental investigation of seismic models for forecasting earthquakes in the northeast region
damage identification using PCA-compressed frequency of India. Bulletin of the Seismological Society of America.
response functions and neural networks. Journal of Sound 105, 6 (Dec. 2015), 2910–2927.
and Vibration. 290, 1–2 (Feb. 2006), 242–263. DOI:https://doi.org/10.1785/0120140361.
DOI:https://doi.org/10.1016/j.jsv.2005.03.016. [27] Smith, L.N. 2016. a Disciplined Approach To Neural
[23] Ozerdem, M.S. et al. 2006. Self-organized maps based Network Hyper-Parameters: Part 1. (2016), 1–21.
neural networks for detection of possible earthquake [28] Srivastava, N. et al. 2014. Dropout: A Simple Way to
precursory electric field patterns. Advances in Prevent Neural Networks from Overfitting.
Engineering Software. 37, 4 (2006), 207–217.
DOI:https://doi.org/10.1016/j.advengsoft.2005.07.004. [29] Su, Y.P. and Zhu, Q.J. 2009. Application of ANN to
prediction of earthquake influence. 2009 2nd
[24] Panakkat, A. and Adeli, H. 2009. Recurrent neural International Conference on Information and Computing
network for approximate earthquake time and location Science, ICIC 2009 (2009), 234–237.
prediction using multiple seismicity indicators. Computer-
Aided Civil and Infrastructure Engineering. 24, 4 (2009),
280–292. DOI:https://doi.org/10.1111/j.1467-
8667.2009.00595.x.
[25] Pasari, S. 2018. Stochastic modelling of earthquake
interoccurrence times in Northwest Himalaya and

View publication stats

You might also like