International Journal of Advances in Engineering and Management (IJAEM)

Volume 4, Issue 1 Jan 2022, pp: 847-852 www.ijaem.net ISSN: 2395-5252

Bitcoin Price Prediction Using Machine

Learning and Neural Network Model
Pratiksha Patil
B.E. Student, Smt. Indira Gandhi College Of Engineering , Ghansoli ,Navi-Mumbai , Maharashtra , India

---------------------------------------------------------------------------------------------------------------------------------------
Submitted: 15-01-2022 Revised: 23-01-2022 Accepted: 26-01-2022
---------------------------------------------------------------------------------------------------------------------------------------
ABSTRACT: Bitcoin has become an investment central bank(1).Because it is accepted in over 40
due to its increasing popularity, and it operates on nations throughout the world , it has prompted the
the Block chain technology, which has given rise to birth of new alternative coins.Other crypto
other crypto currencies.This makes predicting its currencies, products, and services are also
value extremely challenging, thus this predictor is exchanged using Bitcoin(2,3).Due to block chain
put to the test using a Machine Learning Algorithm technology, where each electronic coin is encrypted
and an Artificial Neural Network Model.We used with a unique digital signature that makes it easier
Bitcoin data sets to test and train the machine to track and can be trusted, no hacker has been able
learning and artificial intelligence models in this to breach it since its launch in 2009.Before passing
work.The data filtration process was completed it on, each owner signs a digital hash of the
with the help of python libraries.Python provides a previous transaction and adds the public key of the
fantastic capability for data visualisation and next owner(4,5).Bitcoin's value increased from
analysis.We reduce the data and employ the $1,000 USD in January 2017 to $16,000 USD by
characteristics or attributes that are most suited for the end of December 2017, and it is now worth
the model after we have a good knowledge of it. 32818 USD as of July 2021(6,7).We can say that the
The model is put into action, and the results are crypto market is quite volatile, and Bitcoin, out of
recorded.When compared to other Machine all the crypto currencies available, is the most
Learning models from relevant studies, it was popular among investors due to its anonymity and
determined that the linear regression model's transparency(8,9).The goal of this study is to develop
accuracy rate is extremely high; it was shown to be a Bitcoin price prediction system that employs a
99.87 percent correct.The LSTM model, on the variety of machine learning techniques and deep
other hand, has a 0.08 percent minor error rate.As a learning models.The price of Bitcoin is affected by
result, the neural network model is superior to the a variety of factors; in this project, we will focus on
machine learning model in terms of the open, close, high, and low elements. There are
optimization.The tkinter library was used to six libraries in this work: Pandas , Seaborn , Scikit ,
develop a tiny GUI that allows the user to input the Tikinter , Pickle(10,11).Section 2 of the study article
High, Low, and Open features values and then contains other authors' work that is similar to this
anticipate the coin's future value.The results of a project, Section 3 contains the proposed idea and
machine learning model and an artificial neural technique, Section 4 has a brief summary of the
network model are compared in this research.We data sets and outcomes, and Section 5 contains the
selected linear regression to compare it to the conclusion and future work.
LSTM model because it delivered the highest
accuracy compared to the other machine learning II. RELATED WORK
models. The day trading approach was used to
KEYWORDS: Machine Learning; Artifificial predict the values of ETH, BTC, XEM, XRP,
Neural Network; Bitcoin; Block chain; Crypto XLM, and LTC using the SVM and SVM-PSO
currency. methods, where they used the day trading method
to predict the values of ETH, BTC, XEM, XRP,
I. INTRODUCTION XLM, and LTC.The optimal results are shown by
Bitcoin is a digital crypto currency that SVM-PSO.The accuracy of different Classifiers
runs on a decentralised internet network; it may be varies depending on the coin.However, while this
traded using an online peer-to-peer Bitcoin network article just uses a machine learning technique, the
that is not depending on a single administrator or data might be improved much more by using the

DOI: 10.35629/5252-0401847852 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 847
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 4, Issue 1 Jan 2022, pp: 847-852 www.ijaem.net ISSN: 2395-5252

Deep Learning approach(12).The use of a prices.The experiment yielded a 99.3 percent

transaction graph to anticipate Bitcoin price is accuracy.Although this article has a good accuracy
advocated.The Baseline, Logistic Regression, rate, the data set employed is insufficient for a
SVM, and Neural Network models are used in the model to work on a real-time chart.
experiment, and their accuracy is 53.4 percent, 54.3
percent, 53.7 percent, and 55.1 percent, III. METHODOLOGY
respectively.This paper's feature selection is based Data collection is the first step we do
on the Bitcoin block chain network, which is the when beginning a project.It is defined as the
least informative characteristic for Bitcoin price process of gathering, measuring, and interpreting
prediction(13).On ETH, BTC, and XRP, emotive precise research insights utilising established
analysis and machine learning principles like SVM validation methodologies.An analyst would then be
and Random Forest were used to predict crypto able to evaluate their theory based on the
currency prices, with BTC having the greatest information received.Regardless of the subject of
accuracy of 0.72.Since machine learning study, information collection is by far the most
techniques were used, this accuracy rate has been important and important advance for research. For
quite low, and it can be improved by testing with many disciplines of study, the process for gathering
deep learning models(14).Xgboost Tree, Neural information varies depending on the data
Network, Ensemble Learning Method (with the required.The most important goal of data collecting
best accuracy of 92.4 percent), and KNN model is to ensure that the information acquired is rich in
were used to create a prediction model(15).SVM substance and dependable for statistical analysis,
provides the best accuracy for a time-scale activity allowing for efficient and effective data-driven
consisting of daily, 15-, 30-, and 60-minute returns, decision-making.The data set includes day
according to a prediction system that uses Log transactions from October 24, 2020, to August 15,
regression, SVM, ANN, and random 2020.When there are large or multiple data sets, the
forest.Although SVM tends to produce better data is first verified using regression techniques,
results than the other four algorithms, when Deep and then a deep learning model is deployed to
learning concepts are used, the prediction system provide superior accuracy than machine learning
can still produce better outcomes(16).Using the principles.
open, low, and high costs, a linear regression model
was applied to forecast the various cryptocurrency

Fig 1. Display of the Data collected

Now that we have all of the data we need output faster and more accurately. Data
for the project, we can go on to the following step, visualisation is a technique for better
data segregation or feature selection.Trimming comprehending data or information by presenting it
undesired data or removing unneeded info from a in a diagrammatic format.With the use of graphics,
data set is what this method entails.This phase is data visualisation allows us to communicate data
crucial because we only need the attributes that can relationships.These visuals are in the shape of
help us anticipate the outcome, while additional patterns that are very simple to comprehend.This is
data can produce noise in our final result.To put it one of the primary ways that machine learning aids
another way, we segregate data so that we may with data analysis.Whether you work in finance,
build a better model that gives us an optimum marketing, technical support, or design, you must
outcome, avoid over-fitting or redundancy, and be able to visualise data in order to comprehend
shorten training time so that the system can provide

DOI: 10.35629/5252-0401847852 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 848
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 4, Issue 1 Jan 2022, pp: 847-852 www.ijaem.net ISSN: 2395-5252

it.As a result, data visualisation has become increasingly vital in today's environment
.

Fig 2. The features represented in the data

We can see the link between attributes and between "Volume of USDT" and "Volume of
pinpoint the ones we need with the help of data BTC" with the other features is not evident in the
visualisation libraries.A sample image of the presented image Figure 3, thus they are separated
correlation graph between the features in the given and removed from the training data set.
data set is presented below.The relationship

Fig 3. Correlation graph between the features.

Linear regression is a technique for C = y intercept

determining the relationship between dependent m = slope
and independent variables and predicting future x, y are the points on the graph
events.Simple linear regression is used when only
one dependent and one independent variable are
used.
Multi-linear regression is used when the number of
independent and dependent variables increases.

The graph is drawn using a straight line across the

graph that is calculated using the least square
approach to find the best fit.
y = mx+C

DOI: 10.35629/5252-0401847852 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 849
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 4, Issue 1 Jan 2022, pp: 847-852 www.ijaem.net ISSN: 2395-5252

Long Short-Term Memory, It's a deep accuracy than other algorithms.Unlike prior models
learning notion, specifically a Recurrent Neural based on Hidden Markov Models (HMM) and
Network concept, that avoids the vanishing other comparable concepts, LSTM is capable of
gradient issue.The fundamental advantage of this understanding context-sensitive language.
approach is that it prevents back propagation
mistakes from disappearing or inflating, allowing
them to flow backward via an infinite number of
virtual layers unfurled in space.LSTM mostly
works with time series graphs with data sets
containing events that occurred hundreds or
millions of discrete time steps previously.It can
handle signals with a mix of low and high- V. RESULT AND DISCUSSION
frequency components and works with Following the data analysis, we discovered that
considerable delays between critical only four features were suitable for this project's
incidents.LSTM has been utilised by a number of testing.Only the specified features were remained
researchers to predict time series related data sets after the data was reduced, as shown in Figure 4.
for stock prediction, with greater or higher

Fig 4. Attribute/Features selected are Open, High, Low, and Close

We can see the result of two models: one accuracy of the testing data is also roughly 99.97
is a Machine Learning model, which is Linear percent.Meanwhile, as demonstrated in Figure 6,
regression, and the other is a Recurrent Neural the LSTM model tends to find accuracy with
Network model, which is Long Short-Term Model, respect to the Mean Absolute Error, which displays
which shows us two different outputs. The Mean an error rate of roughly 0.08 percent.The Mean
Squared Equation, which shows us the correctness Squared Equation, which gives us the correctness
of the linear graph with regard to the continuous- of the linear graph with regard to the continuous-
time frame data set, is commonly used in linear time frame data set, is commonly used in
regression.As shown in Figure 5, the accuracy of regression.
the training data is around 99.97 percent, while the

Fig 5. Accuracy obtain from the training and testing data set usingLinear regression model

Fig 6. Testing of Linear regression model

DOI: 10.35629/5252-0401847852 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 850
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 4, Issue 1 Jan 2022, pp: 847-852 www.ijaem.net ISSN: 2395-5252

Fig 7. Final Resultant graph of LSTM and the Mean Absolute Error rate (0.08%)

The data visualisation depicts the [1]. Sin E, Wang L. Bitcoin price prediction
correlation between all of the features, although using ensembles of neural networks. In:
only the four chosen features have a significant 2017 13th International conference on
association.The data is then fitted into the model natural computation, fuzzy systems and
using the specified python commands. knowledge discovery. IEEE. 2017;p.666–
These data models were trained and tested 671. doi:10.1109/FSKD.2017.8393351.
on a small number of data sets, and the results were [2]. Shankhdhar A, Singh AK, Naugraiya S,
provided.With the advancement of technology and Saini PK. Bitcoin Price Alert and Prediction
the expansion of data sets, we may continue to test System using various Models. IOP
the model with a variety of alternative crypto Conference Series: Materials Science and
currencies.The LSTM model has a slightly higher Engineering. 2021;1131(1):012009.
prediction rate than the linear regression model, but Available from:
only by a little margin. https://dx.doi.org/10.1088/1757899x/1131/1/
01200
VI. CONCLUSION [3]. Mittal R, Arora S, Bhatia MP. Automated
According to the findings, Long Short- cryptocurrencies prices prediction using
Term Memory has a higher accuracy rate than machine learning. ICTACT Journal on Soft
Linear Regression.Because this study only Computing. 2018;8(4):1758– 1761.
compares the features of open, close, high, and Available from:
low, the outcome may alter if we examine http://ictactjournals.in/paper/IJSC_Vol_8_Is
additional factors.Data sets cannot be the main s_4_Paper_8_1758_1761.pdf.
rationale for forecasting because the crypto market [4]. Nakamoto S. Bitcoin: A peer-to-peer
is dynamic and influenced by social media and electronic cash system. 2019. Available
other external factors.New data can be acquired, from: https://git.dhimmel.com/bitcoin-
evaluated, and rehearsed as technology progresses, whitepaper/.
resulting in greater findings for this experiment. [5]. Sebastião H, Godinho P. Forecasting and
trading cryptocurrencies with machine
VII. FUTURE WORK learning under changing market conditions.
1) More algorithms are being implemented in Financial Innovation.
order to determine the best approach for [6]. 2021;7(1):1–30. Available from:
predicting the crypto currency. https://dx.doi.org/10.1186/s40854-020-
2) Implementing IOT model for smart automatic 00217-x.
analysis. [7]. Phaladisailoed T, Numnonda T. Machine
3) To work on a better User Interface so that learning models comparison for bitcoin price
people can access these data easily and prediction. 10th International Conference on
effortlessly. Information Technology and Electrical
Engineering. 2018;p. 506–511. Available
REFERENCES from: 10.1109/ICITEED.2018.8534911.

DOI: 10.35629/5252-0401847852 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 851
 International Journal of Advances in Engineering and Management (IJAEM)
Volume 4, Issue 1 Jan 2022, pp: 847-852 www.ijaem.net ISSN: 2395-5252

[8]. Jaquart P, Dann D, Weinhardt C. Short-term bitcoin. 2015.

bitcoin market prediction via machine from:http://snap.stanford.edu/class/cs224w-
learning. The Journal of Finance and Data 2015/
Science. 2021;7:45–66. Available from: projects_2015/Using_the_Bitcoin_Transacti
10.1016/j.jfds.2021.03.001. on_Graph_to_Predict_the_Price_of_Bitcoin.
[9]. Rane PV, Dhage SN. Systematic erudition of pdf.
bitcoin price prediction using machine [16]. 14) Valencia F, Gómez-Espinosa A, Valdés-
learning techniques. 5th International Aguirre B. Price Movement Prediction of
Conference on Advanced Computing & Cryptocurrencies Using Sentiment Analysis
Communication Systems (ICACCS). 2019;p. and Machine Learning. Entropy.
594–598. Available from: 2019;21(6):589. Available from:
10.1109/ICACCS.2019.8728424. https://dx.doi.org/10.3390/e21060589.
[10]. Roy R, Roy S, Hossain MN, Alam MZ, [17]. Chowdhury R, Rahman MA, Rahman MS,
Nazmul N. Study on nonlinear partial Mahdy MRC. An approach to predict and
differential equation by implementing MSE forecast the price of constituents and index
method. Global Scientific of cryptocurrency using machine learning.
[11]. Journals. 2020;8(1):1651–1665. Physica A: Statistical Mechanics and its
[12]. Mckinney W. Pandas: a foundational Python Applications. 2020;551(124569). Available
library for data analysis and statistics. from:
Python for High Performance and Scientific https://dx.doi.org/10.1016/j.physa.2020.1245
Computing. 2011;14(9):1–9. Available from: 69.
https://www.dlr.de/sc/portaldata/15/resource [18]. Akyildirim E, Goncu A, Sensoy A.
s/dokumente/pyhpc2011/submissions/pyhpc Prediction of cryptocurrency returns using
2011_submission_9.pdf. machine learning. Annals of Operations
[13]. Waskom M. Seaborn: statistical data Research. 2021;297(1-2):3–
visualization. Journal of Open Source 36.from:https://dx.doi.org/10.1007/s10479-
Software. 2021;6(60):3021. Available from: 02003575-y.
https://dx.doi.org/10.21105/joss. 03021. [19]. Huang X, Zhang W, Huang Y, Tang X,
[14]. An optimized support vector machine (SVM) Zhang M, Surbiryala J, et al.. 2021.
based on particle swarm optimization (PSO) [20]. Muniye TA, Mohanty MR, Satapathy S.
for cryptocurrency forecasting. Procedia Bitcoin Price Prediction and Analysis Using
Computer Science. 2019;163:427–433. Deep Learning Models. In: Communication
Available from: Software and Networks. 2020. Available
https://doi.org/10.1016/j.procs.2019.12.125. from: 10.1007/978-981-15-5397-4_63.
[15]. Greaves A, Au B. Using the bitcoin
transaction graph to predict the price of

DOI: 10.35629/5252-0401847852 Impact Factor value 7.429 | ISO 9001: 2008 Certified Journal Page 852