A Short Survey on Bitcoin Price Prediction

Stefano Bistarelli1 , Francesco Santini1 and Luca Maria Tutino1,2

1
Dipartimento di Matematica e Informatica, Università degli Studi di Perugia, Via Vanvitelli 1, 06123 Perugia (PG), Italy
2
Dipartimento di Matematica e Informatica “Ulisse Dini”, Università degli Studi di Firenze, Viale Giovanni Battista Morgagni,
67/a, 50134 Firenze (FI), Italy

Abstract
Bitcoin and other cryptocurrencies have emerged as decentralized digital payment systems driven solely by
blockchain technology and supply and demand dynamics. Predicting their prices has become a significant field
of research with important implications in automated trading. This study conducts a Systematic Literature
Review (SLR) to identify the most common features and models used in recent works for Bitcoin price prediction.
By employing an ad hoc method to automate the survey procedure, a summary table is constructed to extract
pertinent information. The objective is to discern the performance of various features and models across different
data ranges and resolutions, laying the groundwork for developing an effective automated trading system. By
synthesizing findings from the literature, this study aims to provide insights into the state-of-the-art methodologies
for cryptocurrency price prediction, facilitating informed decision-making for market agents.

Keywords
Bitcoin, Cryptocurrency, Machine Learning, Price prediction,

1. Introduction
Interest in cryptocurrencies, especially in Bitcoin, has returned to its peak due to the recent bull market,
which has shown even the most skeptical that the price of Bitcoin exhibits a notable regularity linked
to one of the primary (inflationary) characteristics of the algorithm: the periodic (quadrennial) decrease
of the Mining Reward. Despite the difficulty in capturing this periodicity with existing models, it is far
from certain that it does not exist.
Bitcoin [1] is a cryptocurrency created through open-source software operating on peer-to-peer
networks, serving as an irreversible private payment platform. It lacks physical form, is not backed
by any public entity, and thus does not require intervention from government agencies or other
intermediaries like banks to conduct transactions. These transactions occur within the blockchain
system, which acts as an open ledger, efficiently recording transactions between parties and leaving an
indelible mark that cannot be erased. This makes blockchain a decentralized validation protocol that is
difficult to manipulate, with a low risk of fraud.
Being a recently created financial product, Bitcoin still faces the challenge of high volatility (see
Figure 1). According to [2], over seven years, from April 2015 to April 2022, the standard deviation of
Bitcoin’s daily return rate was 3.85%, 2.68 times that of gold’s return rate during the same period, and
3.36 times that of the S&P500 index. Because of these significant price fluctuations, Bitcoin’s function as
a store of value and transactional currency has been questioned, and the findings of several researchers
indicate that Bitcoin might represent a new asset class [3].
While benefiting from BTC’s security and decentralization, understanding its trend to minimize
floating risk poses a complex problem. Apart from this critical issue, owning cryptocurrencies can also
bring high returns to appropriately managed portfolios.
This paper introduces a meticulous Systematic Literature Review (SLR, [4]) focusing on Bitcoin price
prediction. Given the relatively nascent nature of Bitcoin and its unique characteristics, traditional
predictive approaches derived from other asset classes (typically from the stock market) offer only partial

DLT 2024: 6th Distributed Ledger Technologies Workshop, May 14-15 2024, Turin, Italy
$ [email protected] (S. Bistarelli); [email protected] (F. Santini); [email protected] (L. M. Tutino)
€ https://bista.sites.dmi.unipg.it/ (S. Bistarelli); https://francescosantini.sites.dmi.unipg.it/ (F. Santini)
 0000-0001-7411-9678 (S. Bistarelli); 0000-0002-3935-4696 (F. Santini); 0009-0002-5135-9681 (L. M. Tutino)
© 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).

CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
Figure 1: Time evolution of Bitcoin price.

applicability. However, the flexibility of machine learning approaches and the availability of extensive
multidimensional and multimodal data present a promising avenue for forecasting Bitcoin prices. By
undertaking this review, we seek to contribute to the decision-support literature by identifying pertinent
methods, patterns, and promising avenues for further investigation. Additionally, the review aims to
develop robust reporting guidelines to enhance transparency within the field.
Cryptocurrency price prediction is a complex task influenced by various factors such as volatility,
trading volume, mining difficulty, market sentiment, regulatory developments, and technological
advancements. Integrating these diverse factors poses a significant challenge for researchers and
practitioners alike. However, advancements in statistical analysis, machine learning algorithms, and
deep learning techniques offer promising solutions to address these challenges.
The SLR performed in this paper aims to contribute to understanding Bitcoin price prediction by
trying to answer the following research questions:

RQ1: What features are used in the prediction, and from which sources are they taken?

RQ2: What models are used to make the predictions?

RQ3: What models perform best for different data ranges and resolutions?

In Section 2, we report some of the related work in the literature that surveys Bitcoin forecasting;
Section 3 describes the survey methodology step by step, from the choice of the database to the extraction
of the desired information in the final dataset of paper; in Section 4, we give the results analyzing our
records using pies, occurrences histograms, and heat maps. Finally, Section 5 presents the final findings
and ideas about possible future work.

2. Related work
In literature, some recent survey works that have an objective similar to ours, i.e. Bitcoin price prediction,
are: Jaquart et al. 2020 [5]; Khedr et al. 2021 [6]; Patel et al. 2022 [7]; Mezquita et al. 2022 [8]; Fang et al.
2022 [9]; Zhang et al. 2024 [10].
In [5], the authors conducted a literature search following the guidelines proposed by Webster and
Watson [11]. They initiated the process by querying various interdisciplinary research databases.
This search aimed to match their defined search terms in titles, abstracts, or keywords, leading to the
identification of an initial set of 101 publications by April 2019. Each publication’s title and abstract
were then analyzed, resulting in the exclusion of 76 papers that did not explicitly align with the scope
of the literature review. Subsequent forward and backward searches yielded eight additional relevant
articles, bringing the total to 33 papers for in-depth review. The authors then derived key concepts for
paper categorization through an initial screening of 10 recent peer-reviewed conference proceedings
and journal papers, reviewed these papers, and developed a set of initial ideas for classification, which
were refined throughout the screening process. The final synthesis identified four distinct concepts:
Method, Features, Prediction Interval, and Prediction. The authors reflect on the significance of Bitcoin
pricing research via machine learning, highlighting the lack of transparency and comparability across
the reviewed literature. They propose recommendations for future researchers to address these issues,
including structured disclosure of model configurations, publication of models and data to open research
repositories, and benchmarking against other published models.
The contribution of [6] lies in its comprehensive analysis of articles about cryptocurrency price
prediction, spanning the period from 2010 to 2020. It sheds light on the challenges inherent in traditional
approaches and proposes a shift towards Machine Learning (ML) and Deep Learning (DL) paradigms for
more effective solutions. This study offers valuable insights into future research directions and potential
solutions to these challenges, filling a critical gap in existing literature and providing a foundation for
further exploration.
The work in [8] aims to provide a comprehensive overview of the current state of the cryptocurrency
market and the patterns discernible from its activity, explicitly focusing on price prediction. Through a
review of 13 papers, various approaches to predicting cryptocurrency prices are explored, including
analysis of social network and news activity, machine learning algorithms, and market behavior in
response to regulatory actions. The works examined employ diverse methods, such as machine learning
and statistical analysis, to forecast market closing prices and analyze factors like sentiment analysis of
tweets and news, investor reactions to regulatory announcements, and the underlying technology of
each cryptocurrency. It’s concluded that some methods display promise in predicting market prices
accurately and that external factors such as regulations and social sentiment, coupled with pump-and-
dump schemes, complicate prediction tasks.
In [9], the authors give an overview and analysis of the research work on cryptocurrency trading. This
survey presents a vocabulary of the definitions and current state of the art. The paper comprehensively
surveys 146 cryptocurrency trading papers and analyses the research distribution that characterizes the
cryptocurrency trading literature. Research distribution among properties and categories/technologies
is analyzed in this survey. The study also summarises the datasets used for experiments and examines
the research trends and opportunities in cryptocurrency trading. Future research directions and
opportunities are discussed.
The work in [10] provides a comprehensive review of deep learning methods applied in cryptocur-
rency research, spanning price prediction, portfolio management, bubble analysis, abnormal trading
detection, trading regulations, and initial coin offerings. The contribution of this study lies in four key
aspects: conducting a literature review on deep learning models across multiple financial applications,
offering an overview of cryptocurrency history and significant currencies, comprehensively review-
ing deep learning models in cryptocurrency research across diverse tasks, discussing findings from
reviewed studies, and outlining future research directions. The review highlights the popularity of
deep learning models such as convolutional neural networks, recurrent neural networks, deep belief
networks, and deep reinforcement learning in financial applications, particularly for cryptocurrency
price prediction. Researchers have developed LSTM-based derivative models like the SAM-LSTM model,
which outperform traditional prediction models. Deep learning models such as DRL, LSTM, RF, RNN,
GRU, and MLP have also been employed to analyze Bitcoin asset portfolios, detect market bubbles,
identify abnormal trading behaviors, and supervise market transactions. Some studies have also com-
bined multiple deep-learning models to enhance prediction accuracy. Overall, this study contributes to
advancing multidisciplinary research on cryptocurrency and deep learning, providing insights into the
efficacy of deep learning models across various cryptocurrency-related tasks and paving the way for
future research in this domain.
In [7], it is stated that in the realm of cryptocurrency price prediction, researchers deploy a combi-
nation of time-series models like ARIMA and sophisticated deep learning algorithms such as LSTM
and GRU, crucial for capturing the intricate patterns and trends in cryptocurrency price movements,
which are characterized by high volatility and non-linearity. Moreover, comprehensive statistical
analyses of cryptocurrency price histories are imperative to understand these digital assets’ underlying
dynamics and risks. The researchers also pay close attention to market activities, including the actions
of companies, startups, organizations, and influential individuals within the cryptocurrency ecosystem,
as these factors can significantly influence price fluctuations. However, despite the advancements in
predictive modeling, there remains a notable gap in existing research concerning the interdependencies
among different cryptocurrencies. This oversight is addressed by the authors of this study, who propose
a novel fusion model that integrates LSTM and GRU architectures. By considering hierarchical depen-
dencies among cryptocurrencies, their model demonstrates superior predictive accuracy compared to
conventional approaches. Furthermore, the authors emphasize the importance of ongoing research
efforts to enhance prediction accuracy through federated and distributed learning methodologies.

3. Methodology
The first phase of the work was dedicated to developing the review protocol. This part is particularly
critical, as effectively and precisely defining all the steps allows for optimizing and expediting the entire
process. An initial scoping exercise was conducted to glean a preliminary understanding of the existing
research landscape and to ascertain suitable search strategies, such as identifying appropriate databases,
defining time periods, selecting relevant search terms, and considering language restrictions. We then
developed our protocol, structuring the review in steps as follows:

• database selection: selection of the database to use in the analysis;

• string development: development of the string for the database search, using logic operators
and suitable restrictions;
• inclusion criteria: selection of the subset of papers to use in the last phase of the survey;
• extracted information: extraction and analysis of the gathered information.

3.1. Database selection

After the preliminary search, as the first step, we select the database to use in the review. Bibliographic
databases (BDBs) play a vital role in systematic reviews, offering access to standardized descriptions of
documents and advanced query systems and additional services such as citation metrics, bibliometric
data, and links to repositories. The most popular databases for Computer Science research are Google
Scholar1 , Elseviers Scopus2 , and Clarivate Web of Science (WoS3 ), the last two mainly used to ensure
a curated selection of articles. The earliest and the most current content coverage comparisons have
attested that Scopus offers more comprehensive coverage than WoS. In general, it was shown that there
was a significant overlap between the information indexed by Scopus and WoS, with Scopus, even if
the overlap between was determined to vary significantly across disciplines like pointed out in [12],
indexing more unique sources than WoS [13].
Elsevier’s Scopus contains peer-reviewed scientific output in books, journal articles, conference
proceedings, and more than 90 million abstracts, bibliometric data, and bibliographic citations. Aca-
demic subjects covered include health sciences (medicine, veterinary, professions health); life sciences
(biology, agriculture, biochemistry, pharmacy) = 17%; physical sciences (chemistry, physics, mathemat-
ics, engineering, information technology, social sciences) = 27%; and social sciences and humanities
(art, literature, psychology, economics, and social sciences) = 31%. Leveraging its advanced search
features, Scopus allowed us to narrow our search parameters effectively and efficiently. Additionally,
the platform’s extensive database afforded us greater confidence in the reliability and relevance of the
articles retrieved.

1
Scholar: https://scholar.google.com/
2
Scopus: https://www.scopus.com/
3
WoS: https://www.webofscience.com/
3.2. String development
Following the database selection, we worked on creating the search string on Scopus. This component
must be carefully curated to balance finding a broader pool of publications and methodically choosing
a smaller set of papers that satisfy inclusion criteria.
Our search string consisted of the three main key terms: "Bitcoin", "Price", and "Prediction". Based on
our research objectives, these terms were complemented with alternatives using the OR logic operator
and asterisks to include all the forms of the words prediction and forecasting. We also requested
our papers to be written in English, to be articles or conference papers, to be published before 2024,
and to belong to the four subject areas of Computer Science, Engineering, Mathematics and Economics,
Econometrics, and Finance. Additionally, we decided to limit our search to the titles of the paper to limit
the number of results.
The specific query used was:
TITLE ( ( c r y p t o c u r r e n c y OR b i t c o i n ) AND p r i c e AND ( p r e d i c t ∗ OR f o r e c a s t ∗ ) )
AND PUBYEAR < 2 0 2 4 AND ( LIMIT −TO ( SUBJAREA , "COMP" ) OR LIMIT −TO ( SUBJAREA ,
" ENGI " ) OR LIMIT −TO ( SUBJAREA , " DECI " ) OR LIMIT −TO ( SUBJAREA , "MATH" ) OR
LIMIT −TO ( SUBJAREA , "ECON" ) ) AND ( LIMIT −TO ( LANGUAGE , " E n g l i s h " ) ) AND
( LIMIT −TO ( DOCTYPE , " cp " ) OR LIMIT −TO ( DOCTYPE , " a r " ) )

The search produced 372 articles since 2017, an initial dataset that will be further refined in the next
phase of our analysis.
Starting from this dataset of articles, we generated a csv table using an integrated tool in Scopus,
which will be the basis for subsequent analyses. The fields of the table are:

• Authors: the authors of the paper;

• Title: the title of the paper;
• Year: year of publication;
• Cited by: number of citation;
• Source Title: name of the publishing journal;
• DOI: Digital Object Identifier of the paper;
• Link: URL of the article;
• Abstract: brief summary of the paper;
• References: references of the article.

The table was then imported into a Colab Notebook to automate the various survey phases as much
as possible and extract useful information in histograms and pie charts.
Figure 2 depicts the number of articles published annually within the timeframe considered by our
survey; the research trend is generally upward, albeit with irregular growth, partly due to the extreme
volatility of Bitcoin and other cryptocurrency prices. The first paper is by Jang et al., which dates back
to 2017 [14]).
The distribution of categories among articles and conference papers is almost equally portioned, 49%
of the articles being Conference papers, while the rest articles.
Figure 3 shows the distribution of citation numbers through the dataset of papers, from which we
notice that the majority of documents, with few exceptions barely visible in the histogram, have a low
number of citations.
Figure 4 shows the distribution of categories among the different subjects, Computer Science being
the top one, followed by Engineering and Mathematics.

3.3. Inclusion criteria

Due to the large number of articles, we have decided to focus our more in-depth analyses on a subset of
approximately 100 articles, aiming to consider the most relevant ones, especially those from the last
three years, 2021-2023, the period most pertinent to our analysis.
Figure 2: Number of papers in the initial database for each year in the time interval our survey considers.

Figure 3: Distribution of the citations.

We used the articles’ citations, crafting a normalized index to consider mostly more recent and
relevant articles, given by:
C
S=
(2024 − y)2
that is a score 𝑆 proportional to the number of citations 𝐶 (as counted 8 Feb 2024), and inversely
proportional to the square of the difference between 2024 (now) and the year 𝑦 of publication of the
considered paper. In this way, we eliminated articles without citations from the second part of our
analysis. We also excluded surveys that did not develop their model and papers that did not include
BTC price as one of the study targets.
Figure 4: Distribution of subjects.

Figure 5: Number of selected and discarded papers to form the dataset for the final analyses.

Figure 5 shows the number of selected and discarded articles year by year. We notice that, especially
in 2022 and 2023, with our inclusion criteria, discarded articles are way more than the selected ones. In
the future, we will consider more papers for the final dataset.

3.4. Extracted information

The analysis aims to compile and organize some of the data on Bitcoin price prediction in the literature.
To do this, we created a second table by starting with the first and manually adding additional fields
for each of the 102 articles chosen. The method takes time since it takes a lot of reading to obtain the
necessary information from all this material.
The added fields (with the legenda), shown in tables 1 and 2 in the Appendix A, are:

• Crypto: objective (dependent) variables of the prediction (i.e., price of BTC, ETH, etc.);
Figure 6: The distribution of the objective of the price prediction (dependent variable).

• Features: independent variables (features) used for the prediction;

• Models: models (algorithms) used for the prediction;
• Evaluation metrics: metrics used to evaluate the performance of the models for Bitcoin price
prediction;
• Data range: time interval analyzed in the considered paper;
• Data resolution: resolution(s) of the model (es. 1d, 1 min, etc.);
• Source: online source where historical data are retrieved.

Manually inserted data must be standardized to add records for later suitable statistical analysis intel-
ligently. Appending a legend listing all the traits and models examined is crucial. Many articles utilize
ensemble models, which are combinations, and in some cases hybrid (in the sense that they combine
statistical methods with machine learning), of various models to obtain predictions. However, this
possibility is not considered in our analysis as we currently aim only to understand, through statistical
analysis, the most commonly used models without considering the various hybrid combinations. In
this case, the configurations would be numerous, and the analysis would fail to extrapolate almost any
regularity.

4. Results
In the final stage of the review, we try to answer the research questions by extracting the desired
information from the definitive set comprising approximately 100 papers described at the end of the
previous section. We imported this second refined dataset into the Colab notebook, where all the survey
steps are implemented and automated. This allowed us to import our dataset, initially in CSV format,
convert it into a pandas data frame, and interact with it more efficiently, given the numerous tools
available in this Python environment. We then created various graphs.
Figure 6 shows the percentage of articles predicting only Bitcoin and those predicting Bitcoin and
other cryptocurrencies. Keeping in mind that we discarded papers that did not predict the price of
Bitcoin, it shows that the vast majority of the considered papers only predict the price of BTC. At
the same time, the rest also predict other crypto prices, like ETH, LTC, and XRP, which are primarily
cryptocurrencies with high capitalization.
Figure 7a displays the list and occurrences of the 25 most frequently used features. As expected, we
notice that the most utilized features are the candle data (OHCL) and the volume of transactions V [15],
followed by Market Capitalization and some on-chain features like Hash Rate and Difficulty [16]. Also,
macroeconomic indexes like S&P500, the price of metals like gold or other materials [17], the price of
other cryptocurrencies (ETH, LTC, etc., [18]), and many price-related technical indicators like MACD
 (a) Occurrences of predictive features. (b) Occurrenceses of predictive models.

Figure 7: Occurrences of predictive features and models architecture.

Figure 8: Evolution of the top 4 most used architectures throughout time.

and RSI are used for the forecasting of Bitcoin price [19]. In the works that utilize Sentiment Analysis
(SA), Google Trends Index and Twitter Sentiment Score are mostly considered [20]. Figure 7b shows
the occurrences of the top prediction model architectures. Deep learning models based on LSTM and
GRU are the most widely used [21, 22], followed by the machine learning techniques, SVM and RF [23],
and by the ARIMA time series forecasting statistical method [24].
Figure 8 shows the evolution of the top 4 most used architectures through time. We notice that
LSTMs are always at the first position and that machine learning algorithms, like RF and SVM, have
re-gained prominent positions in 2023. The same applies to the ARIMA statistical forecasting method,
often used as a benchmarking model.
The metrics used to evaluate the performance of the price prediction depend on the paper’s approach,
that is, whether the forecasting is given in the form of regression (R) or classification (C). In the former
case, metrics like MSE, RMSE, MAE, and MAPE are used, while in the latter, Accuracy, Precision, Recall,
and F1-score are used. From figure 9a we can see that the RMSE is the most frequent used metric,
suitable to regression models, followed by MSE and MAPE. For classification tasks, the most used metric
is the accuracy.
Figure 9b shows the error ranges for the models used in 2023 papers. It has to be pointed out that
it is complicated to compare different models that give predictions for different dataset ranges and
 (a) Most Occurring evaluation metrics. (b) Error ranges of the models.
Figure 9: Figure 9a shows the occurrences of the evaluation metrics used in 2023 papers, while figure 9b shows
the error ranges of some regression models.

resolutions, so we decided to provide error ranges for the most current models to give an idea of their
effectiveness.

(a) Most Occurring Data Resolution. (b) Top 20 Most Occurring Data Sources

Figure 10: Most Occurrent Data Resolutions and Sources.

Figure 10a and Figure 10b show respectively the top occurrent data resolutions and sources. The most
frequent forecasting time interval is one day, and the most popular site for retrieving financial data is
coinmarketcap4 . In contrast, for Sentiment Analysis, which uses textual datasets, the most common
source site is X (i.e., Twitter5 ).
Figure 11 shows the dataset ranges’ heat map. We notice that the datasets begin in 2009, the genesis
year of blockchain technology and Bitcoin, and have a peak at the beginning of 2018, right after the
first massive adoption phase, when the capitalization of the Crypto market reached for the first time
the level of 1 T$.

5. Conclusion
The primary contribution of our paper lies in the recent analysis of articles about Bitcoin price prediction,
encompassing both traditional statistical and machine learning techniques. Extracting insights from a

4
Coinmarketcap: https://coinmarketcap.com
5
Twitter: https://twitter.com/
Figure 11: Heat-map of the datasets ranges.

dataset of articles published between 2017 and 2023, we observe the nascent stage of the cryptocurrency
domain, evident from most papers emerging in the past three years. This observation suggests fertile
ground for future research endeavors in this domain. By consolidating relevant papers and analyzing
their contributions, we aim to lay a foundational framework to support forthcoming research in Bitcoin
price forecasting.
Despite the inherent challenges posed by the volatile nature of cryptocurrencies and their suscepti-
bility to various internal and external factors, researchers have endeavored to develop robust prediction
models. Time-series-based models, such as ARIMA and other regression-based approaches, have tra-
ditionally been employed for price prediction due to their reliance on historical price data. With the
advent of deep learning algorithms, researchers have integrated these advanced techniques into their
methodologies, utilizing variants such as RF, SVM, LSTM, GRU, and hybrid or ensemble models. These
efforts signify a continuous evolution in pursuing more accurate and reliable cryptocurrency price
forecasts.
In the future, we plan to leverage the information gathered from the present survey to define and
assemble a multimodal (financial and textual) dataset and design and implement our portfolio of efficient
predictors. The final goal is to build an advanced Automatic Trading System to optimize the management
of Crypto assets and support short—and long-term investments.

Acknowledgments
The authors are members of the Gruppo Nazionale Calcolo Scientifico-Istituto Nazionale di Alta
Matematica (GNCS-INdAM) and of Consorzio Cini. This work has been partially supported by:

• GNCS-INdAM, CUP_E53C22001930001 and CUP_E53C23001670001;

• European Union - Next Generation EU PNRR MUR PRIN - Project J53D23007220006 EPICA:
“Empowering Public Interest Communication with Argumentation”;
• University of Perugia - Fondo Ricerca di Ateneo (2020, 2021, 2022) - Projects
BLOCKCHAIN4FOODCHAIN, FICO, AIDMIX, “Civil Safety and Security for Society”;
• European Union - Next Generation EU NRRP-MUR - Project J97G22000170005 VITALITY: “Inno-
vation, digitalisation and sustainability for the diffused economy in Central Italy”;
• Piano di Sviluppo e Coesione del Ministero della Salute 2014-2020 - Project I83C22001350001 LIFE:
“the itaLian system Wide Frailty nEtwork” (Linea di azione 2.1 “Creazione di una rete nazionale
per le malattie ad alto impatto” - Traiettoria 2 “E-Health, diagnostica avanzata, medical devices e
mini invasività”);
 • project “SERICS” (PE00000014) under the MUR National Recovery and Resilience Plan funded by
the European Union - NextGenerationEU.

References
[1] S. Nakamoto, Bitcoin: A peer-to-peer electronic cash system, Decentralized business review
(2008).
[2] J. Chen, Analysis of bitcoin price prediction using machine learning, J. Risk Financial Management
(2023).
[3] F. Glaser, K. Zimmermann, M. Haferkorn, M. Weber, M. Siering, Bitcoin - asset or currency?
revealing users’ hidden intentions, ECIS 2014 Proceedings - 22nd European Conference on
Information Systems (2014).
[4] B. Kitchenham, O. Pearl Brereton, D. Budgen, M. Turner, J. Bailey, S. Linkman, Systematic literature
reviews in software engineering – a systematic literature review, Information and Software Tech-
nology 51 (2009) 7–15. URL: https://www.sciencedirect.com/science/article/pii/S0950584908001390.
doi:https://doi.org/10.1016/j.infsof.2008.09.009, special Section - Most Cited Arti-
cles in 2002 and Regular Research Papers.
[5] P. Jaquart, D. Dann, C. Martin, Machine learning for bitcoin pricing - A structured literature
review, in: N. Gronau, M. Heine, H. Krasnova, K. Poustcchi (Eds.), Entwicklungen, Chancen und
Herausforderungen der Digitalisierung: Proceedings der 15. Internationalen Tagung Wirtschaftsin-
formatik, WI 2020, Potsdam, Germany, March 9-11, 2020. Zentrale Tracks, GITO Verlag, 2020, pp.
174–188.
[6] A. M. Khedr, I. Arif, M. El-Bannany, S. M. Alhashmi, M. Sreedharan, Cryptocurrency price
prediction using traditional statistical and machine-learning techniques: A survey, Intelligent
Systems in Accounting, Finance and Management 28 (2021) 3–34.
[7] N. P. Patel, R. Parekh, N. Thakkar, R. Gupta, S. Tanwar, G. Sharma, I. E. Davidson, R. Sharma,
Fusion in cryptocurrency price prediction: A decade survey on recent advancements, architecture,
and potential future directions, IEEE Access 10 (2022) 34511–34538.
[8] Y. Mezquita, A. B. Gil-González, J. Prieto, J. M. Corchado, Cryptocurrencies and price prediction:
A survey, Lecture Notes in Networks and Systems 320 LNNS (2022) 339 – 346. Cited by: 9.
[9] F. Fang, C. Ventre, M. Basios, L. Kanthan, D. Martinez-Rego, F. Wu, L. Li, Cryptocurrency trading:
a comprehensive survey, Financial Innovation 8 (2022). Cited by: 121; All Open Access, Gold Open
Access, Green Open Access.
[10] J. Zhang, K. Cai, J. Wen, A survey of deep learning applications in cryptocurrency, iScience 27
(2024) 108509.
[11] J. Webster, R. T. Watson, Analyzing the past to prepare for the future: Writing a literature review,
MIS quarterly (2002) xiii–xxiii.
[12] M. Visser, N. J. Van Eck, L. Waltman, Large-scale comparison of bibliographic data sources: Scopus,
web of science, dimensions, crossref, and microsoft academic, Quantitative science studies 2 (2021)
20–41.
[13] P. Wouters, M. Thelwall, K. Kousha, L. Waltman, S. De Rijcke, A. Rushforth, T. Franssen, P. Wouters,
et al., The metric tide, Literature review. Supplementary report I to the independent review of the
role of metrics in research assessment and management (2015).
[14] H. Jang, J. Lee, An empirical study on modeling and prediction of bitcoin prices with bayesian
neural networks based on blockchain information, IEEE Access 6 (2018) 5427–5437.
[15] M. J. Hamayel, A. Y. Owda, A novel cryptocurrency price prediction model using gru, lstm and
bi-lstm machine learning algorithms, AI (2021).
[16] S. Ranjan, P. Kayal, M. Saraf, Bitcoin price prediction: A machine learning sample dimension
approach, Computational Economics (2023).
[17] J. W. Goodell, S. Ben Jabeur, F. Saâdaoui, M. A. Nasir, Explainable artificial intelligence modeling
to forecast bitcoin prices, International Review of Financial Analysis 88 (2023) 102702.
[18] Y. Zhu, J. Ma, F. Gu, J. Wang, Z. Li, Y. Zhang, J. Xu, Y. Li, Y. Wang, X. Yang, Price prediction of
bitcoin based on adaptive feature selection and model optimization, Mathematics (2023).
[19] L. Cocco, R. Tonelli, M. Marchesi, Predictions of bitcoin prices through machine learning based
frameworks, PeerJ Computer Science (2021).
[20] Z. Zhou, Z. Song, H. Xiao, T. Ren, Multi-source data driven cryptocurrency price movement
prediction and portfolio optimization, Expert Systems with Applications 219 (2023) 119600.
[21] V. Kumar T, S. Santhi, K. G. Shanthi, G. M, Cryptocurrency price prediction using lstm and recurrent
neural networks, in: 2023 2nd International Conference on Applied Artificial Intelligence and
Computing (ICAAIC), 2023, pp. 1–5.
[22] A. Dutta, S. Kumar, M. Basu, A gated recurrent unit approach to bitcoin price prediction, Juornal
of Risk and Financial Management (2020).
[23] N. Maleki, A. Nikoubin, M. Rabbani, Y. Zeinali, Bitcoin price prediction based on other cryptocur-
rencies using machine learning and time series analysis, Scientia Iranica (2020).
[24] M. Poongodi, V. Vijayakumar, N. Chilamkurti, Bitcoin price prediction using arima model, Inter-
national Journal of Internet Technology and Secured Transactions 10 (2020) 396–406.
[25] A. A. Oyedele, A. O. Ajayi, L. O. Oyedele, S. A. Bello, K. O. Jimoh, Performance evaluation of
deep learning and boosted trees for cryptocurrency closing price prediction, Expert Systems with
Applications 213 (2023) 119233.
[26] X. Sun, M. Liu, Z. Sima, A novel cryptocurrency price trend forecasting model based on lightgbm,
Finance Research Letters 32 (2020) 101084.
[27] P. L. Seabe, C. R. B. Moutsinga, E. Pindza, Forecasting cryptocurrency prices using lstm, gru, and
bi-directional lstm: A deep learning approach, Fractal and Fractional (2023).
[28] Z. Chen, C. Li, W. Sun, Bitcoin price prediction using machine learning: An approach to sample
dimension engineering, Journal of Computational and Applied Mathematics 365 (2020) 112395.
[29] S. McNally, J. Roche, S. Caton, Predicting the price of bitcoin using machine learning, in: 2018
26th Euromicro International Conference on Parallel, Distributed and Network-based Processing
(PDP), 2018, pp. 339–343.
A. Tables and legenda

Paper Prediction Obj. Features Models Ev. Metrics

[25] BTC, ETH, O, H, L, C, V, SMA, ADAB, GBM, XGB, MAPE = 1-3%
BNB, LTC, EMA, WP MLP, GRU, CNN
XLM, DOGE
[26] BTC, 41 other DJ, S&P500, HangSengI, LightGBM, SVM, Accuracy = 48-62%
cryptos USDFI, ShanghaiStockI, RF
ShenzhenCompI, RMB,
FTSEChinaA50, Oil, O,
H, L, C, V
[27] BTC, ETH, LTC O, H, L, C, AdjC LSTM, GRU, Bi- RMSE = 1031-1280;
LSTM MAPE = 0,036-0,057
[28] BTC BlockSize, H-R, Diff, LSTM, SVM, RF, Accuracy = 48-66%,
NumberTrx, ConfTrx, XGB, LR Precision =49-72%,
MempoolTrxCount, Recall = 35-74%, F1-
MempoolSize, Mar- score = 48-78%
ketCap, EstTrxValue,
TrxFees, GoogleTI,
Gold
[29] BTC O, H, L, C, Diff, H-R LSTM, RNN, Accuracy=50-52%,
ARIMA RMSE=5-53%

Table 1
Part 1: First five columns

Dataset range Dataset resolution Source

01/01/2018 - 1d yahoo.com,
02/03/2022 bitfinex.com,
investing.com
01/01/2018 - 1d investing.com
30/06/2018
01/01/2018 - 1d yahoo.com
01/01/2023
02/02/2017 - 1 d, 5 min coinmarketcap.com
01/02/2019
19/08/2013 - 1d coindesk.com,
19/07/2016 blockchain.info

Table 2
Part 2: Last two columns.

O: open price of the considered interval; H: max price; L: min price; C: closing price; V: volume of
transactions; H-R: Hash rate; Diff: mining difficulty; NumberTrx: number of transactions; ConfTrx:
confirmed transactions; MempoolTrxCount: Mempool transaction count; TrxFees: transaction
fees; GoogleTI: Google Trends Index; DJ: Dow Jones Index; MACD: Moving Average Convergence
Divergence; RSI: Relative Strength Index.