Prediction of Earthquake Using Machine

Learning Algorithms
Gaurav Singh Manral Alka Chaudhary
Amity Institute of Information Technology Amity Institute of Information Technology
Amity University Noida, Uttar Pradesh Amity University, Noida, Uttar Pradesh
gaurav.manral@s.amity.edu achaudhary4@amity.edu

ABSTRACT - An earthquake is one of the most This study intends to investigate the application of
devastating natural catastrophes that may cause machine learning methods for earthquake
major infrastructure damage and casualties. prediction, more especially the Random Forest
Early earthquake detection can be crucial for Regressor and Neural Network Model. Using
minimising damage and saving lives. Machine parameters like longitude, latitude, and timestamp
learning is a powerful tool that may be used to data, the study focuses on forecasting the size and
forecast earthquakes based on historical seismic depth of earthquakes. The dataset utilised in this
data and other geographical data. This study study contains earthquake recordings from diverse
examines the viability of predicting earthquakes areas, and both mean absolute error and mean
using machine learning methods, especially the squared error measures are used to assess how well
Random Forest Regressor and Neural Network the models performed.
Model. The purpose of this study is to make
earthquake magnitude and depth predictions This study seeks to increase earthquake forecast
utilising factors including time, place, and precision, which might have substantial effects on
previous seismic activity data. disaster planning and response initiatives.
Predictions that are accurate can lessen earthquake
Keywords- Machine Learning Algorithms, damage and save lives. This work sheds light on
Dataset, Random Forest Regressor, Neural the potential of these models for forecasting the
Network, predictions. size and depth of earthquakes. The application of
machine learning techniques in earthquake
I. NTRODUCTION prediction is a relatively new field of research.
1.1 Earthquake II. LITERATURE REVIEW
An earthquake is a natural occurrence that happens It provides an overview of recent research on
when the Earth's crust suddenly moves or shifts. earthquake prediction using machine learning
Buildings, infrastructure, and other structures are techniques, including neural networks, support
frequently damaged as a result of the seismic waves vector machines, and decision trees [1] (magnitude
that are produced by this movement. The most estimation not included). This article examines the
severe earthquakes cause extensive devastation and use of machine learning algorithms for earthquake
human casualties, while their intensity and length prediction, including the application of artificial
can vary. Many things, including as tectonic neural networks and fuzzy logic [2] (Not applied
activity, volcanic eruptions, and man-made other algorithms).
operations like mining or drilling, can cause them.
This other article summarizes recent research on
1.2 Predicting Earthquakes earthquake prediction using machine learning
Natural catastrophes such as earthquakes have the algorithms, including random forests, deep
potential to seriously harm both human life and learning, and convolutional neural networks [3]
infrastructure. It is difficult to accurately estimate (not provided comprehensive evaluation). This
the size and depth of earthquakes since this requires survey article provides an overview of recent
examining intricate data from several sources. research on earthquake prediction using machine
Machine learning algorithms have made significant learning techniques, including artificial neural
strides recently, and this holds tremendous networks, support vector machines, and deep
potential for earthquake prediction accuracy. learning [4]. (does not provide a quantitative
evaluation).
This study compares the performance of various strongly influenced by the quality of the training
machine learning algorithms for earthquake data. Pre-processing data is therefore necessary to
prediction, including decision trees, support vector guarantee that the data is accurate, reliable, and
machines, and artificial neural networks [5] pertinent to the issue at hand.
(Limited dataset).
Data cleaning, which entails deleting any blank or
III. EXPERIMENT AND METHODOLOGY incorrect numbers, managing outliers, and
resolving discrepancies in the data, is the initial
A. Dataset stage of data pre-processing. Before the data is
The dataset used to forecast earthquakes using utilised to train a machine learning model, this is
machine learning methods like Random Forest done to make sure that it is correct and
Regressor and Neural Network Model often comprehensive.
comprises data on a variety of previous To further enhance the quality and usefulness of the
earthquakes. The following details are contained in data, data transformation also entails the
the dataset: application of methods including dimensionality
1. Date and Time: when the earthquake took reduction, feature engineering, and data
place. augmentation. These methods can enhance the
model's functionality, decrease the complexity of
2. Location: The earthquake's epicentre’s the data, and make the model more generalizable to
latitude and longitude. new data.
3. Magnitude: The earthquake's strength as In conclusion, data preparation is a crucial stage in
measured by the Richter scale. machine learning that entails preparing data for use
in an algorithm by cleaning, formatting, and
4. Depth: The earthquake's epicentre depth.
modifying it. The quality and effectiveness of the
5. Further features: Other characteristics, machine learning model can be considerably
such as the kind of fault that triggered the enhanced by ensuring that the data is accurate,
earthquake, the separation from the next reliable, and pertinent to the issue being handled.
fault, and the distance from the closest
shoreline, may be important for
earthquake prediction. C. PROPOSED METHOD
i)Random Forest Regressor
The dataset has to be big enough for machine
learning models to be able to learn from the data An effective machine learning method for
and generate precise predictions. To prevent the predicting earthquakes is random forest regression.
models from being biased towards a specific sort of This particular ensemble learning technique makes
earthquake, the data should also be representative predictions by combining different decision trees.
of earthquakes that happen in many places and at
various times. A training dataset is used by the algorithm in a
random forest regression model for earthquake
The dataset is available from a number of sources, prediction to learn the correlation between several
including government organisations that monitor characteristics, such as latitude, longitude, depth,
earthquakes, academic institutions that gather and timestamp, and the related earthquake
earthquake data, and internet databases that magnitudes and depths.
compile earthquake data from numerous sources.
To prevent the findings from being skewed, it is The system creates several decision trees, each of
crucial to verify that the data is accurate and which makes an earthquake prediction based on a
trustworthy and that any duplicates or outliers be subset of the data, including the depth and
eliminated from the dataset. magnitude. To guarantee that the decision trees are
varied and minimise overfitting, the algorithm
B. Pre-processing randomly chooses a portion of the features and a
subset of the training data to utilise for each
Data cleaning, formatting, and transformation are
decision tree.
essential steps in the pre-processing phase of
machine learning, which prepares data for use in a By averaging the forecasts of all the decision trees
machine learning algorithm. The accuracy and in the forest, the model may be used to predict
performance of a machine learning model are fresh data once it has been trained. Moreover, the
system may produce significant feature relevance In order to forecast earthquake depth and
ratings that can be used to determine which features magnitude layer by layer, this suggested technique
have the strongest correlations with earthquake combines the capabilities of both random forest
depth and magnitude. regression and neural network models, which can
increase earthquake prediction's precision and
ii) Neural network Model dependability. To tune the models' hyperparameters
A neural network is a type of machine learning and assess the models' effectiveness using various
model that mimics the structure and function of the datasets, more study is necessary.
human brain. It consists of a network of D. Experiment
interconnected nodes, or neurons, which work
together to process data and form hypotheses. One Step 1: Importing various libraries needed to create
or more characteristics make up the network's the model and dataset.
input. These features are then converted into a set
of output values through a sequence of weighted
connections and nonlinear activation functions.
Using methods like backpropagation, the network
modifies the weights of the connections between
neurons during training to reduce the discrepancy
between the expected and actual output values. The
use of neural networks in a range of applications,
including as speech and picture recognition, natural
language processing, and predictive modelling, is
proving to be a useful tool for tackling challenging
challenges.

In the situation mentioned above, the linear

regressor was more of a kind where the projected Step 2: Determine the key characteristics from the
values were not what was anticipated. Therefore, seismic data, namely the Date, Time, Latitude,
we are now building the neural network to suit the Longitude, Depth, and Magnitude, and then build
training set's data. Three dense layers, each with an object with those characteristics.
sixteen nodes, two nodes, and the activation
functions relu, relu, and softmax, make up a neural
network.

A neural network with three dense layers is a

specific form of neural network design made up of
three dense layers—or layers of completely linked
neurons—of neurons. The network may learn
intricate nonlinear correlations between the input
Step 3: Visualizing data
and output data because every neuron in a dense
layer is linked to every neuron in the layer above it. Here, a globe map with all of the earthquakes from
the database is visualised, providing a clear
The input data is supplied into the first dense layer depiction of the regions where earthquake
of a three dense layer neural network, which frequency will be higher.
consists of a certain number of neurons, each with a
set of learnable weights and biases. The second
dense layer receives the output from the first dense
layer as input, and so on for the third dense layer.

The output of each dense layer's neurons is

subjected to a nonlinear activation function, which
enables the network to learn increasingly intricate
representations of the input data. Rectified Linear
Unit (ReLU), sigmoid, and tanh are frequently used
activation functions.
 b. Neural Network Model

Step 4: Data Splitting

The data should first be divided into X’s and y’s,
which correspond to the model's input and output,
respectively. Here, the outputs are magnitude and
depth, with time stamp, latitude, and longitude as
the inputs. To train and test with validation, divide
the X’s and y’s. There are 80% of the training
dataset and 20% of the test dataset. IV. CONCLUSION

To conclude, this paper is implemented Random

Forest Regressor and neural network algorithm to
predict the Earthquake with parameters as time,
place, and previous seismic activity data. As per
the results both models can do the predictions with
the dataset.
Step 5: Building the models and predicting the machine learning is a useful technique for
score. predicting earthquakes. It is possible to estimate the
a. Random Forest Regressor likelihood of earthquakes by analysing historical
seismic data and other geospatial data using
methods like random forests and neural networks.
These methods can be combined with sensors and
other monitoring systems to provide early warning
of future earthquakes and lessen the impact of this
catastrophic natural disaster with more
development and improvement.

A long period of research has been devoted to the

complicated and difficult topic of earthquake
prediction. Neural networks and random forest
regressors are two examples of machine learning
methods that have produced promising earthquake
prediction outcomes. A type of machine learning
technique called a neural network may discover
patterns and connections in huge datasets. They can
examine large data sets and spot tiny trends that
may be hard for people to notice, making them well
suited for seismic assessment. Using neural
networks and random forest regressors, earthquake
prediction has a bright future. Machine learning
algorithms will be able to examine more data as it
becomes available and will be better able to
anticipate when and where earthquakes will most
likely occur. The ability to better prepare for
earthquakes and lessen their effects on people and
infrastructure will be very helpful to emergency
personnel. Machine learning methods can be
utilized for various earthquake-related activities
besides prediction, like real-time earthquake
detection and seismic activity monitoring. This can
aid researchers in better comprehending the origins
and consequences of earthquakes and in coming up
with innovative solutions to lessen their
effects .Overall, neural networks and random forest
regressors have great promise for the future of
earthquake prediction, and in the years to come,
earthquake research and disaster planning are
expected to heavily rely on these machine learning
methods.

REFRENCES
1."Earthquake prediction using machine learning: a review" by
L. Huang and M. Wyss, was published in the Geophysical
Journal International in 2019.

2. "Application of machine learning in earthquake prediction"

by M. E. Yetginer, et al., was published in the International
Journal of Disaster Risk Reduction in 2020.

3. "Earthquake prediction using machine learning algorithms: a

comprehensive review" by S. K. Singh and K. Singh, in the
Journal of Seismology (2021).

4. "Earthquake prediction using machine learning: a survey" by

S. M. S. Karim and A. H. M. Kamruzzaman, in the Journal of
Earthquake Engineering (2021).

5. "Earthquake prediction using machine learning algorithms: a

comparative study" by H. R. Karimi and M. R. Jahed-Motlagh,
in the Journal of Seismology (2021).