2023 International Conference on Data Science and Its Applications (ICoDSA)

A Systematic Literature Review: Forecasting Stock

Price Using Machine Learning Approach
Niclauss Lumoring Daniel Chandra Alexander Agung Santoso Gunawan
Mathematic Department Mathematic Department Computer Science Department
School of Computer Science School of Computer Science School of Computer Science
Bina Nusantara University Bina Nusantara University Bina Nusantara University
2023 International Conference on Data Science and Its Applications (ICoDSA) | 979-8-3503-0519-7/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICoDSA58501.2023.10277318

Jakarta, Indonesia 11480 Jakarta, Indonesia 11480 Jakarta, Indonesia 11480

[email protected] daniel.chandra @binus.ac.id [email protected]

AbstractÏWith the increasing popularity of stock trading, recognize stock patterns from vast amounts of data representing
individuals, and financial entities such as investment companies, underlying stock prices.
hedge funds, and retail investors are actively participating in the Predicting the stock market is a complex task influenced by
stock market to generate profits. Numerous strategies, ranging numerous factors. Before investing in a stock, individuals
from traditional methods employing fundamental and technical typically conduct two types of analysis. The first is fundamental
analysis to modern approaches utilizing cutting-edge
analysis, where investors consider factors such as the intrinsic
technology, have been developed and implemented. However,
determining the optimal method remains challenging. Designing value of stocks, market and economic performance, political
an effective strategy in the complex and dynamic stock market environment, and more to determine whether to invest. The
environment poses significant difficulties. Therefore, this study second is technical analysis, which involves analyzing market
aims to provide an overview of machine learning applications in data, such as past prices and volumes, to evaluate equities.
the stock market and identify the most frequently used machine Technical analysts focus on stock charts to identify patterns and
learning models or methods for market prediction. trends that can predict future stock behavior. Investing in the
Furthermore, the study aims to identify the strategy that stock market requires extensive knowledge and information, as
achieves the highest accuracy in predicting stock prices. Based there are many factors that can affect stock prices, including a
on a systematic literature review encompassing quantitative and
company's past performance, analyst opinions, and the
qualitative analyses, it is found that Support Vector Machine
(SVM) emerges as the most popular machine learning technique effectiveness of its strategies [2].
for predicting stock prices. However, Long Short-Term The primary method of stock market prediction involves
Memory (LSTM) stands out as the machine learning technique forecasting the future value of individual stocks, market
that demonstrates the highest level of accuracy, achieving an segments, or the entire market. Many professionals and experts
impressive accuracy rate of 99.58%. These findings highlight the
in machine learning, artificial intelligence, trading, data analysis,
effectiveness of LSTM in forecasting stock values with
exceptional precision. The results of this study contribute to a
and computer engineering have dedicated their efforts to this
comprehensive understanding of machine learning in the stock area. The performance of a company's revenue and sales, which
market domain and offer valuable insights for developing can be influenced by factors such as regulatory constraints,
effective trading strategies. Traders and investors can leverage microeconomic indicators, supply, and demand, greatly affects
LSTM and SVM techniques to enhance their stock market the value of its shares. Investing in the market entails various
predictions and make informed investment decisions. market risks. The stock market is continuously influenced by
KeywordsÏmachine learning, stock market price, prediction societal sentiments, and traditional financial and economic
model analysis methods are often insufficient for accurately predicting
future stock market prices. To forecast future stock trends, the
I. INTRODUCTION fundamental analysis method utilizes quantitative analysis of
Stock market forecasting has traditionally relied on stock prices, inflation, interest rates, and other variables [3].
conventional methods, such as analyzing technical and Several learning approaches are used to anticipate stock
fundamental aspects of stock market prices. However, with values, including Long Short-Term Memory (LSTM),
the emergence of machine learning applications, stock market Autoregressive Integrated Moving Average (ARIMA), and
forecasting has become more accessible and precise. Various Convolutional Neural Network (CNN). As technology advances,
machine learning techniques have been employed to predict trading becomes more accessible, and artificial intelligence and
stock market prices. The equity capital market serves as a machine learning play vital roles. Machine learning algorithms
platform for issuing and trading shares of listed companies. have significantly advanced in the business sector. Stock traders
As stocks represent partial ownership in a company and offer employ algorithms to expedite and simplify trading, utilizing
resources and security, the stock market provides investors artificial intelligence to adapt to various trade patterns.
with a marketplace to buy and sell these investable assets [1]. Algorithmic trading is inevitably evolving into useful machine
Predicting the direction of future stock prices, rather than learning.
their exact value, is crucial for financial decision-making. By Currently, there are numerous methods available for
being aware of the expected direction of a stock, investors can predicting stock trends. While traditional regression methods
determine whether to buy or sell. Moreover, research has were initially used for this purpose, technological advancements
shown that predicting direction rather than value can enhance have enabled the utilization of various analytical techniques in
financial success. Machine learning (ML) techniques are stock market prediction. Examples include artificial neural
particularly popular nowadays due to their ability to networks (ANN) and support vector machines (SVM), each

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE SAO PAULO. Downloaded on March 25,2024 at 21:58:18 UTC from IEEE Xplore. Restrictions apply.
979-8-3503-0519-7/23/$31.00 ©2023 IEEE 129
 2023 International Conference on Data Science and Its Applications (ICoDSA)

employing unique methods for learning patterns and This systematic literature review utilized the PRISMA
forecasting the stock market. checklist approach to evaluate research papers on forecasting
stock prices using machine learning. The study selection process
By automating the utilization of existing static data,
involved:
machine learning advancements can be leveraged to predict
financial markets [4]. Although stock price movements are Identification: A total of 18,100 records were identified through
typically nonlinear, researchers can employ linear forecasting a comprehensive database search, and an additional 10 records
models to anticipate short-term changes in stock prices based were found through other sources.
on historical data. Machine learning programs can learn from Screening: After combining the records, 18,050 were screened
historical data to identify patterns and generate desired based on predefined criteria.
outcomes. Eligibility: Out of the screened records, 50 full-text articles were
identified as potentially meeting the eligibility criteria.
Therefore, the aim of this paper is to explore the role of
machine learning in stock market prediction, as well as Exclusion: After assessing the full-text articles, 20 were
identify the most popular and accurate machine learning excluded due to factors such as not meeting inclusion criteria or
models. The study follows a structured literature review, poor methodology.
which is arranged as follows: first we discuss the Inclusion: A total of 30 studies were included in the qualitative
methodology of our paper in section 2, and then discuss the synthesis, where the findings were analyzed and summarized.
results in section 3. Finally, we conclude our paper on section By following the PRISMA checklist, this systematic literature
4 with suggestions for further research. review ensures transparency and rigor in the selection and
evaluation of relevant research papers for the topic of interest.
II. METHODOLOGY
A. Research Questions
This literature paper utilizes a systematic literature review
(SLR) methodology, which involves formulating three This research conducts a SLR to achieve the aim to review
research questions and collecting relevant data from the relevant work of machine learning approach in stock market
literature. The objective of this literature review is to price prediction.
investigate machine learning approaches for stock market RQ1: What does machine learning in stock market price
prediction, specifically focusing on the most used approaches prediction look like in general?
and their performance scores. To achieve this, the paper poses RQ2: What is the frequently used model/approaches of ML in
research questions related to machine learning approaches in stock prediction?
stock prediction and provides answers based on the findings
RQ3: Which technique is more effective in terms of accurately
obtained from the literature review.
predicting stock prices?
The PRISMA approach for SLR was used to evaluate the
research paper used in this study. The PRISMA flowchart is B. Criteria Research Paper
shown as Fig. 1 below. For this literature review, the selection criteria include
research papers and articles written in English, such as journals,
research papers, and literature reviews, specifically focused on
Identification

Record identified Additional records the domain of "machine learning approach on the stock market."
through database identified through The publication year of the papers considered is between 2017
search n = 18,100 other sources n = 10
and 2022, and the title of the research paper is considered for
screening.
Initially, a total of 18,100 publications containing the
keyword "machine learning technique on the stock market" were
identified through Google Scholar. After an initial screening
Screening

Records screened
n = 18050 process, 18,050 papers were rejected due to reasons such as
duplication, irrelevance to the study objective, or mismatch with
the research topic. This filtering process resulted in 50 papers
that were considered for further review.
In the subsequent review, 19 papers were disqualified after a
more detailed assessment. Finally, based on the screening and
Eligibility

Numbers of Full text articles

records eligibility record excluded n review process, the conclusion of this literature review is based
full text article n = 20 on 31 research studies that met the inclusion criteria and
= 50
provided relevant insights into the topic of machine learning
approaches in the stock market.
III. RESULT AND DISCUSSION
Studies paper RQ1 Î What does machine learning in stock market price
Included

included in prediction look like in general?

qualitative
synthesis n = 30 In recent decades, the stock market has experienced
significant transformation due to technological advancements,
with machine learning playing a crucial role in various aspects
such as predicting stock prices, stock returns, and stock market
Fig.1 Prisma flowchart. behavior. This development has had a positive impact on

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE SAO PAULO. Downloaded on March 25,2024 at 21:58:18 UTC from IEEE Xplore. Restrictions apply.
130
 2023 International Conference on Data Science and Its Applications (ICoDSA)

industry professionals by providing more accurate stock TABLE II. METRIC RESULT TO EACH PAPER
market predictions for investors. Machine learning has Methods Metrics Study
revolutionized the process of stock market prediction by Identifiers
enabling the forecasting of market behavior, stock prices, and DDPG
Standard Error:
[7]
introducing new models for data processing and market 13.62%
analysis. SVM:
Accuracy: 65.2%
Numerous studies have focused on the accuracy of stock SVM, LTSM [17]
market price forecasting and have proposed various LTSM:
approaches. However, the rapid advancement of technology Accuracy: 66.5%
has outpaced the growth in this field. Currently, there are four SVM:
Accuracy: 78.7%
main categories that describe the recent advancements in SVM, Random
[24]
stock analysis and forecasting: sentiment analysis, machine Forest
Random Forest:
learning, pattern recognition, and statistical analysis. These Accuracy: 80.8%
categories encompass a range of techniques and algorithms SVM:
used for analyzing stock market data and making predictions. Accuracy: 68.2%
Recall: 65.2%
It is worth noting that some machine learning algorithms Precision: 64.7%
integrate techniques from both technical analysis and F-Score: 64,9%
fundamental research to forecast stock markets. However,
most of these approaches fall under the broader category of Random Forest:
technical analysis [1]. Accuracy: 80.7%
Recall: 78.3%
RQ2 Î What is the frequently used model/approaches of SVM, Random Precision: 75.2%
ML in stock prediction? Forest, KNN, F-Score: 76,7%
[29]
Logistic KNN:
Each of the research papers has a different approach to Regression Accuracy: 65.2%
forecast stock prices using machine learning approaches. Recall: 63.6%
Precision: 64.8%
Table I shows what method in machine learning F-Score: 64,1%
approaches is the most used to forecasting stock prices.
Logistic Regression:
TABLE I. METHOD TO FORECASTING STOCK PRICE Accuracy: 78.6%
Recall: 76.6%
Number of Precision: 77.8%
Methods Study Identifiers
Papers F-Score: 77,1%
Artificial Neural [9], [10], 19], Root Mean Square
6 LTSM [31]
Networks (ANN) [25], [26], [27] Error: 3.69%
[13], [15], [17], Differential
Support Vector [20], [23], [24], Mean Error: 4.17%
10 Privacy (DP) - [32]
Machine (SVM) [25], [26], [27], Accuracy: 99.58%
LTSM
[29]
Artificial Intelligence 1 [25]
[17], [23], [27], Table II presents the accuracy results for various machine
Long Short-Term
7 [28], [30], [31],
Memory (LTSM)
[32]
learning techniques in predicting stock prices, with a particular
[5], [16], [19], focus on Differential Privacy (DP) - Long-Short Term Memory
Random Forest 7 [21], [24], [27], (LSTM). DP is one of the components used in modeling LSTM,
[29] and its purpose is to express patterns within a dataset while
K-nearest Neighbor 3 [21], [27], [29] concealing individual-specific information. This is achieved by
Logistic Regression 2 [21], [29] introducing random noise, such as from a Laplace or Gaussian
Deep Deterministic distribution, which compromises the exact accuracy of the result
Policy Gradient 1 [7]
(DDPG) and masks the specific contents of each row. The DP-LSTM
technique achieved an outstanding accuracy of 99.58%.
In addition to DP-LSTM, other effective machine learning
According to Table I, the method most often employed is techniques for predicting stock prices are Random Forest and
SVM. Ten studies use SVM. LTSM and Random Forest are Support Vector Machine (SVM). Random Forest achieved an
the second and third most well-liked machine learning accuracy of 80.8%, while SVM achieved an accuracy of 78.7%.
methods for forecasting stock values, respectively, with only These techniques, although not as accurate as DP-LSTM, still
a small gap between them (7 publishing of papers). demonstrate significant predictive capability in forecasting stock
RQ3 Î Which technique is more effective in terms of prices.
accurately predicting stock prices?
IV. CONCLUSION
Each of the research papers has a different approach to
forecast stock prices using machine learning approaches. This paper encompasses a compilation of various studies that
Table II shows what method in machine learning approaches explore the application of machine learning in predicting stock
produces the highest accuracy of forecasting stock prices. prices. The main objective of this paper is to identify different
machine learning methods and assess their accuracy in stock
price forecasting. Based on the findings from the literature
review, the most popular machine learning technique for
predicting stock prices is SVM. However, the machine learning
technique that demonstrates the highest level of accuracy in

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE SAO PAULO. Downloaded on March 25,2024 at 21:58:18 UTC from IEEE Xplore. Restrictions apply.
131
 2023 International Conference on Data Science and Its Applications (ICoDSA)

forecasting stock values is LSTM, achieving an impressive [14] N. Rouf, M. B. Malik, T. Arif, S. Sharma, S. Singh, S. Aich, and H. C.
accuracy rate of 99.58% [32]. It is important to note that while Kim, ÐStock market prediction using machine learning techniques: a
LSTM exhibits near-perfect accuracy, traders should not rely decade survey on methodologies, recent developments, and future
solely on this method due to the inherent risks associated with directions,Ñ Electronics, vol. 10, p. 2717, 2021.
trading. Market conditions and situations are dynamic and
[15] V. K. S. Reddy and K. Sai, ÐStock market prediction using machine
subject to change. Therefore, traders should exercise caution
learning,Ñ International Research Journal of Engineering and
and consider multiple factors when making trading decisions.
Technology (IRJET), vol. 5, pp.1033-1035, 2018.
Nonetheless, this paper aims to provide valuable insights to
assist traders and simplify the trading process. [16] S. Kompella and K. C. C. C. Chilukuri, ÑStock market prediction using
machine learning methods,Ñ International Journal of Computer
Engineering and Technology, vol. 10, p. 2019, 2020.
REFERENCES
[17] N. Arora, ÐFinancial analysis: Stock market prediction using deep
[1] D. Shah, H. Isah, and F. Zulkarnaen, ÐStock market analysis: A
learning algorithms,Ñ In Proceedings of International Conference on
review and taxonomy of prediction techniques,Ñ International
Sustainable Computing in Science, Technology and Management
Journal of Financial Studies, vol. 7, p. 26, 2019.
(SUSCOM), Amity University Rajasthan, Jaipur-India, 2019.
[2] A. Thakkar and K. Chaudhari,ÐCREST: Cross-Reference to
[18] S. Vadlamudi, ÐStock Market Prediction using Machine Learning: A
Exchange-based Stock Trend Prediction using Long Short-Term
Systematic Literature Review,Ñ American Journal of Trade and Policy,
MemoryÑ, Procedia Computer Science, vol. 167, pp. 616-625,
vol. 4, pp. 123-128, 2017.
2020.
[19] B. Weng, ÐApplication of machine learning techniques for stock
[3] S. Yao, ÐHigh-Frequency Stock Trend Forecast Using LSTM
market prediction,Ñ 2017.
Model.Ñ 2018 13th International Conference on Computer Science
[20] S. Kamley, S. Jaloree, and R. S. Thakur, ÐPerformance forecasting of
& Education (ICCSE), 2018.
share market using machine learning techniques: A review,Ñ
[4] S. A. de Ara›jo, D. F. de Barros, E. M. da Silva, et al, ÐApplying
International Journal of Electrical and Computer Engineering, vol. 6,
computational intelligence techniques to improve the decision
p.3196, 2016.
making of business game players,Ñ Soft Comput 23, pp. 8753Î
[21] N. Bhardwaj and M. A. Ansari, ÐPrediction of stock market using
8763, 2019.
machine learning algorithms,Ñ International Research Journal of
[5] Cervell–-Royo, R. and Guijarro, F. (2020) ÐForecasting stock
Engineering and Technology, vol. 6, 2019.
market trend: A comparison of machine learning algorithms,Ñ
[22] M. Umer, M. Awais, and M. Muzammul, ÐStock Market Prediction
Finance, Markets and Valuation, 6(1), pp. 37Î49. Available at:
Using Machine Learning AlgorithmsÑ, ADCAIJ: Advances in
https://doi.org/10.46503/nluf8557.
Distributed Computing and Artificial Intelligence Journal, vol. 8, pp.
[6] Zhang, J. and Lei, Y. (2022) ÐDeep Reinforcement Learning for
97Î116, 2019.
stock prediction,Ñ Scientific Programming, 2022, pp. 1Î9.
[23] S. K. Lakshminarayanan and J. P. McCrae, ÐA Comparative Study of
Available at: https://doi.org/10.1155/2022/5812546.
SVM and LSTM Deep Learning Algorithms for Stock Market
[7] Liu, X.Y. et al. (2018) Practical Deep Reinforcement Learning
Prediction,Ñ In AICS, pp. 446-457, 2019.
Approach for Stock Trading, pp. 1Î7.
[24] M. M. Akhtar, A. S. Zamani, S. Khan, A. S. A. Shatat, S. Dilshad, and
[8] Kabbani, T. and Duman, E. (2022) ÐDeep Reinforcement Learning
F. Samdani, ÐStock market prediction based on statistical data using
Approach for trading automation in the stock market,Ñ SSRN
machine learning algorithms,Ñ Journal of King Saud University-
Electronic Journal [Preprint]. Available at:
Science, vol. 34, p.101940, 2022.
https://doi.org/10.2139/ssrn.4100283. J. Clerk Maxwell, A Treatise
[25] T. J. Strader, J. J. Rozycki, T. H. Root, and Y. H. J. Huang,Ñ Machine
on Electricity and Magnetism, 3rd ed., vol. 2. Oxford: Clarendon,
learning stock market prediction studies: Review and research
1892, pp.68Î73.
directions,Ñ Journal of International Technology and Information
[9] M. Hiransha, E. A. Gopalakrishnan, V. K. Menon, and K. P.
Management, vol. 28, pp.63-83, 2020.
Soman, ÑNSE stock market prediction using deep-learning
[26] C. L. Cocianu and H. Grigoryan, ÐMachine Learning Techniques for
models,Ñ Procedia computer science, vol. 132, pp.1351-1362,
Stock Market Prediction,Ñ A Case Study of Omv Petrom. Economic
2018.
Computation & Economic Cybernetics Studies & Research, vol. 50,
[10] X. Zhong and D. Enke, ÐPredicting the daily return direction of the
2016.
stock market using hybrid machine learning algorithms,Ñ Financial
[27] J. Sen and T. D. Chaudhuri, ÐStock price prediction using machine
Innovation, vol. 5, pp.1-20, 2019.
learning and deep learning frameworks,Ñ In Proceedings of the 6th
[11] M. Leippold, Q. Wang, and W. Zhou, ÐMachine learning in the
International Conference on Business Analytics and Intelligence,
Chinese stock market. Journal of Financial Economics,Ñ vol. 145,
Bangalore, India, pp. 20-22, 2018.
pp.64-82, 2022.
[28] S. Kim, S. Ku, W. Chang, and J. W. Song, ÐPredicting the direction of
[12] M. Nabipour, P. Nayyeri, H. Jabani, S. S. and A. Mosavi,
US stock prices using effective transfer entropy and machine learning
"Predicting Stock Market Trends Using Machine Learning and
techniques,Ñ IEEE Access, 8, pp.111660-111682, 2020.
Deep Learning Algorithms Via Continuous and Binary Data; a
[29] A. Pathak and S. Pathak, ÐStudy of machine learning algorithms for
Comparative Analysis," in IEEE Access, vol. 8, pp. 150199-
stock market prediction,Ñ International Journal of Engineering
150212, 2020.
Research & Technology, pp.2278-0181, 2020.
[13] V. H. Shah, ÐMachine learning techniques for stock prediction,Ñ
[30] A. Moghar and M. Hamiche,Ñ Stock market prediction using LSTM
Foundations of Machine Learning| Spring, vol. 1, pp.6-12, 2007.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE SAO PAULO. Downloaded on March 25,2024 at 21:58:18 UTC from IEEE Xplore. Restrictions apply.
132
 2023 International Conference on Data Science and Its Applications (ICoDSA)

recurrent neural network,Ñ Procedia Computer Science, 170,

pp.1168-1173, 2020.
[31] J. Sen, S. Mehtab, A. and Dutta, ÐStock price prediction using
machine learning and LSTM-based deep learning models.Ñ 2021.
[32] Li, X. et al., ÐDP-LSTM: Differential Privacy-inspired LSTM for
Stock Prediction Using Financial News, pp. 1Î9, 2019.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE SAO PAULO. Downloaded on March 25,2024 at 21:58:18 UTC from IEEE Xplore. Restrictions apply.
133