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

net/publication/378008818

Stock Market Price Prediction using Machine Learning Techniques

Article in American International Journal of Sciences and Engineering Research · February 2024
DOI: 10.46545/aijser.v7i1.308

CITATION READS

1 1,982

1 author:

Mahfuz Islam Khan Jabed

The Millennium University
1 PUBLICATION 0 CITATIONS

SEE PROFILE

All content following this page was uploaded by Mahfuz Islam Khan Jabed on 07 February 2024.

The user has requested enhancement of the downloaded file.

 AMERICAN INTERNATIONAL JOURNAL OF SCIENCES AND ENGINEERING RESEARCH 7(1) (2024), 1-6

Sciences and Engineering Research

AIJSER VOL 7 NO 1 (2024) E-ISSN 2641-0311 P-ISSN 2641-0303
Available online at www.acseusa.org
Journal homepage: https://www.acseusa.org/journal/index.php/aijser
Published by American Center of Science and Education, USA

STOCK MARKET PRICE PREDICTION USING MACHINE

LEARNING TECHNIQUES

Mahfuz Islam Khan Jabed (a)1

(a)
Student, Department of CSE, The Millennium University, Dhaka, Bangladesh; E-mail: [email protected]

ARTICLE INFO ABSTRACT

Article History: Predicting stock market prices is a challenging task in the financial sector, where the Efficient Market
Hypothesis (EMH) posits the impossibility of accurate prediction due to the inherent uncertainty and
Received: 1st October 2023 complexity of stock price behaviour. However, introducing Machine Learning algorithms has shown the
Reviewed & Revised: 2nd October feasibility of stock market price forecasting. This study employs advanced Machine Learning models
2023 to 30th January 2024 that can predict stock price movements with the right level of accuracy if the correct parameter tuning
Accepted: 1st February 2024 and appropriate predictor models are developed. In this research work, the LSTM model, which is a
Published: 3rd February 2024 type of Recurrent Neural Network (RNN), time series forecasting Facebook Prophet algorithm and
Random Forest Regressor model have been implemented on 10 Dhaka Stock Market (DSEbd) listed
Keywords: companies and six international giants for predicting the stock and forecasting the future price. The
dataset of domestic companies is extracted from the graphical representation of the DSEbd website, and
LSTM, Facebook Prophet, Random the international companies' dataset is imported from Yahoo Finance. In this experiment, Facebook
Prophet demonstrates a long period of forecasting with reasonable accuracy, capturing daily, weekly,
Forest Regressor, Machine Learning,
and yearly seasonality, including holiday effects for market trend analysis. Remarkably, the LSTM model
RNN. exhibits significant accuracy, yielding the best results with evaluation metrics, including RMSE (0.35),
MAPE (0.50%), and MAE (0.30). The experimental results underscore the efficiency of LSTM for future
JEL Classification Codes : stock forecasting, observed over 15 days of upcoming market prices. A comparison of the results shows
that the LSTM model efficiently forecasts the next day's closing price.
H54, P42, G17, C88
© 2024 by the authors. Licensee ACSE, USA. This article is an open access articledistributed under
the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativec
ommons.org/licenses/by/4.0/).

INTRODUCTION
The stock market constantly fluctuating in the fast-paced financial sector, finding it challenging to formulate reliable stock
predictions. Investors and traders feel compelled to seek out creative approaches to reduce investment risks and maximize
profits because of the possibility of generating substantial profits. Machine learning (ML) is becoming a strong tool for
intelligent investment management that can help investors do better at buying and selling stocks. It can also be used to make
decisions and handle portfolios in new ways. This research investigation investigates three distinct models: Facebook
Prophet, which is intended for time series forecasting and focuses on weekly, yearly, and seasonal trends; Long Short-Term
Memory (LSTM), a type of Recurrent Neural Network that can forecast values in the future by learning patterns from
sequential data; and the Random Forest Regressor model from Ensemble Learning, that develops multiple decision trees
during training to improve overall accuracy and generalization. This study distinguishes between model performances using
standard assessment metrics, such as Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), and Root
Mean Square Error (RMSE), using historical data from ten domestic and six worldwide companies: Amazon, Apple,
Microsoft, Google, Netflix, and Tesla. The ultimate objective is to help organizations estimate stock requests and improve
future strategies. The study's findings indicate the Facebook Prophet algorithm's ability to give prolonged predicting periods
with reasonable accuracy. For market trend research, it also records seasonality on a daily, weekly, and yearly scale,
encompassing holiday impacts. In the meanwhile, LSTM shows accurate forecasting for the majority of company stock
prices with careful parameter adjustment for each dataset with a 15-day future prediction observation. The best MAPE is

1
Corresponding author: ORCID ID: 0009-0001-2141-2894
© 2024 by the authors. Hosting by ACSE. Peer review under responsibility of American Center of Science and Education, USA.
https://doi.org/10.46545/aijser.v7i1.308

To cite this article: Jabed, M. I. K. (2024). STOCK MARKET PRICE PREDICTION USING MACHINE LEARNING TECHNIQUES. American
International Journal of Sciences and Engineering Research, 7(1), 1–6. https://doi.org/10.46545/aijser.v7i1.308
1
 Jabed, American International Journal of Sciences and Engineering Research 7(1) (2024), 1-6

produced by LSTM, which is better to other models according to previous study findings. Over the past few decades, there
has been a fundamental improvement in information technology that has changed the path that organizations take. Financial
markets, one of the most fascinating innovations, have a direct impact on the economy of the country (Hiransha et al., 2018).
The global stock market value has topped 68.654 trillion US dollars, according to a (World Bank, 2018) study.

LITERATURE REVIEW
For the purpose of predicting stock prices, a number of deciding algorithms, such as fuzzy systems, decision trees, neural
networks, and genetic algorithms, have been developed. There is a significant relationship between public mood and pattern
discovery, which makes pattern identification an important tool for stock market trend prediction. Previous research on price
forecasting has looked into combining artificial neural networks (ANN) with technical indicators. In the realm of data
mining, Association Rule Mining (ARM) is a well-studied method. Despite this, decision trees are acknowledged for being
useful in financial decision-making. Stock market data prediction is based on a Neural Network Approach in research by
Yoon and Swales (1991), including both quantitative and qualitative elements for decision-making. With a 77.5% accuracy
rate, the four-layer deep neural network beats the conventional MDA model; nevertheless, because of hidden units, it has
trouble interpreting the importance of input parameters. The Autoregressive Integrated Moving Average (ARIMA) model
is used for time series forecasting in another research by Pai and Lin (2005). In order to overcome these difficulties, the
study uses neural networks and Support Vector Machines (SVM) in recognition of ARIMA's limits in capturing nonlinear
patterns. (Akhtar et al., 2022) investigates machine learning algorithms, such as Random Forest and SVM, for stock market
prediction, emphasizing dataset preprocessing, introducing a predictive model for stock lifespan, and achieving an 80.3%
accuracy with the Stock Prediction algorithm.

MATERIALS AND METHODS

Data Collection
The historical data of 10 companies listed on the Dhaka Stock Market has been manually extracted from the Dhaka Stock
Exchange (DSE, 2018) official website. This dataset covers a 6-month period. Additionally, data for 6 international giants’
companies has been collected from yahoo Finance, spanning a 10-year period. The dataset includes information such as
Open, High, Low, Close, Adjacent Close, and Volume for the international companies, as well as Price and Date columns
for the 10 domestic companies. Table 1 and Table 2 display the statistic of international and domestic companies’ datasets,
respectively, along with their training and testing periods.

Table 1. Time Period of International Companies dataset

Dataset Training Dataset Testing Dataset

Time Period 01/01/2012 – 21/09/2023 01/01/2012 – 19/05/2021 20/05/2021 – 21/09/2023

Table 2. Time Period of Domestic Companies dataset

Dataset Training Dataset Testing Dataset

Time Period 02/02/2023 – 01/08/2023 02/02/2023 – 01/07/2023 02/07/2023 – 01/08/2023

Additional Data Included

Three new data columns have been generated, encompassing moving averages (MA) over varying timeframes. These data
columns include 5-day, 10-day, and 15-day MA for domestic companies and 10-day, 20-day, and 60-day MA for
international companies.

Long Short-Term Memory (LSTM Model)

The Long Short-Term Memory (LSTM) model is a sophisticated data mining technique designed to identify fundamental
trends and analyse complex patterns in unstructured data. LSTM model is implemented as a neural network composed of
LSTM units, each consisting of a cell state, input gate, forget gate and output gate. These essential components
collaboratively interact with the input layer, incorporating features such as historical stock prices, date/time information,
and relevant market indicators. The input gate determines the information to be stored in the long-term memory cell, the
forget gate decides what to retain or discard from the past, and the output gate shapes the final output based on the current
input and stored information. This dynamic interaction enables LSTMs to effectively capture and retain crucial patterns and
dependencies in sequential data, establishing their suitability for predicting stock market prices. The specific number of
LSTM units is subject to variation based on task complexity and dataset characteristics, often determined through iterative
experimentation and turning in the model development process.

2
 Jabed, American International Journal of Sciences and Engineering Research 7(1) (2024), 1-6

Figure 1. Proposed Recurrent Neural Network (LSTM)

Facebook Prophet
Facebook Prophet, a powerful time series predictor, identifies nonlinear trends, including seasonal and holiday effects. Ideal
for datasets with strong seasonal results and limited history, Prophet stands out in handling missing data and trend shifts. It
ensures quick and accurate forecasts with minimal computation time, making it equivalent to Stan models. The methodology
involves data collection, pre-processing and input structuring with “ds” for data and “y” for the target variable. The model
initialized with holidays, fits to the training dataset, generating and evaluating predictions. Parameter tuning balances
flexibility, and the user-friendly final model is deployed for real-world predictions, accommodating various “human”
seasons.

Random Forest Regressor

Random Forest Regressor methodology involves collecting historical time series data, pre-processing for cleanliness, and
structuring it with relevant features and a target variable. The model is then initialized with parameters and trained to learn
patterns. Feature importance analysis highlights key variables, followed by validation and hyperparameter tuning.
Predictions are made on unseen data, and performance is evaluated using regression metrics. Visualization aids
interpretation, and upon satisfactory results, the model is deployed for real-world predictions. The Random Forest Regressor,
recognized for its robustness, proves effective in accurate time series regression.

RESULTS AND DISCUSSIONS

To evaluate the model’s forecasting accuracy and overall efficiency, a comparison is conducted among three techniques
(LSTM, PROPHET, RANDOM FOREST) across 10 Dhaka Stock Exchange-listed companies, namely Agrani Insurance
Co. Ltd, Apex Footwear Limited, Berger Paints Bangladesh Ltd, Delta Life Insurance Company Ltd, Deshbandhu Polymer
Limited, Dhaka Insurance Limited, Eastern Insurance Company Ltd, Jamuna Oil Company Limited, Premier Bank Ltd,
Unilever Consumer Care Limited and six international giants, including Amazon, Apple, Google, Microsoft, Netflix and
Tesla. To ascertain the accuracy of the predicted closing prices, Metrics such as Root Mean Square Error (RMSE), Mean
Absolute Percentage Error (MAPE) and Mean Absolute Error (MAE) are utilized to assess errors and accuracy.
RMSE is calculated utilizing equation 1.

∑𝑛 ̂ 𝑖 )2
𝑖=1(𝑦𝑖 −𝑦
RMSE = √ (1)
𝑛

Where: n is the number of observations, yi is the actual value of the observation and 𝑦̂𝑖 is the predicted value. RMSE
is a metric to measure the average accuracy of a predicted model by quantifying the differences between predicted and
observed values.
3
 Jabed, American International Journal of Sciences and Engineering Research 7(1) (2024), 1-6

MAPE is calculated utilizing equation 2.

1 (𝑦𝑖 −𝑦̂𝑖 )
MAPE = ∑𝑛𝑖=1 ∗ 100 (2)
𝑛 𝑦𝑖

Where: n is the number of data points, yi is the actual value and 𝑦̂𝑖 is the forecasted value. MAPE (Mean Absolute
Percentage Error) is used to measure the accuracy of a forecasting method by calculating the percentage difference between
predicted and actual values.

MAE is calculated utilizing equation 3.

1
MAE = ∑𝑛𝑖=1( 𝑦𝑖 − 𝑦̂𝑖 ) (3)
𝑛

Where: n is the number of observations, yi is the actual value and 𝑦̂𝑖 is the predicted value. MAE (Mean Absolute
Error) is used to measure the average magnitude of errors between predicted and actual values, providing a straightforward
assessment of prediction accuracy. The effectiveness of these models is evaluated by providing a 15-day forecast of closing
prices. In the course of this research, the forecasted prices for the next 15 days are compared with the actual closing prices
observed after 15 days when new data becomes available. The analysis is based on the notable accuracy achieved in
predicting the stock prices, where the forecasted closing price exhibit significant similarity with the real data. This
comparison serves as a validation of the models’ predictive capabilities.

Eastern Insurance Company Ltd Stock Jamuna Oil Company Ltd Stock

Netflix Stock Tesla Stock

Figure 2. Original v/s Predicted Stock Closing Price using LSTM Algorithm

Apex Footwear Limited Stock Delta Life Insurance Company Ltd Stock

4
 Jabed, American International Journal of Sciences and Engineering Research 7(1) (2024), 1-6

Google Stock Apple Stock

Figure 3. Original v/s Predicted Stock Closing Price using Facebook Prophet Algorithm

Figure 2 displays the predicted closing prices of stocks for two domestic companies and two international
companies using LSTM. In Figure 3, graphs compare the original closing prices of stocks with their predicted values using
Facebook Prophet. Table 3 presents a comprehensive analysis of RMSE, MAPE and MAE values obtained from LSTM,
Facebook Prophet and Random Forest Regressor model. The results emphasize the strong predictive performance of LSTM
in forecasting stock prices, showcasing its effectiveness in the studied context

Table 3. Comparative Analysis of the RMSE, MAPE and MAE values derived from the utilization of LSTM, Facebook
Prophet and Random Forest Regressor.

LSTM Facebook Prophet Random Forest Regressor

Company Name RMSE MAPE MAE RMSE MAPE MAE RMSE MAPE MAE
Agranins 1.83 1.54% 1.44 2.53 3.65% 3.85 2.21 3.33% 3.13
ApexFoot 1.36 1.28% 1.78 2.64 3.30% 4.69 3.83 3.60% 3.94
BargerPBL 2.98 1.91% 2.97 1.60 1.46% 2.05 1.59 1.05% 1.84
DeltaLife 1.82 1.68% 1.84 3.26 5.19% 3.24 1.59 1.22% 1.34
DeshBandhu 1.21 1.26% 1.55 2.73 2.63% 3.55 2.36 1.74% 2.17
Dhakains 1.71 1.63% 1.79 2.79 3.40% 2.08 1.53 1.47% 1.28
Easternins 2.11 2.74% 2.97 3.31 3.48% 5.61 2.78 1.89% 2.71
JamunaOil 1.48 0.65% 1.16 2.18 2.89% 2.58 1.20 1.09% 1.17
PremierBan 0.35 0.50% 0.30 0.63 1.18% 1.02 1.03 1.09% 1.02
UnileverCL 0.11 0.78% 1.06 1.24 2.65% 2.34 1.99 2.02% 3.63
Amazon 1.96 2.50% 1.71 1.43 1.97% 2.38 1.42 1.05% 1.66
Apple 1.96 1.46% 1.31 2.49 3.21% 5.07 1.37 1.08% 1.08
Google 1.51 2.36% 1.85 1.46 1.80% 3.32 1.20 1.74% 1.86
Microsoft 3.49 2.34% 2.88 1.51 1.94% 2.18 2.22 2.62% 1.69
Netflix 1.72 1.00% 1.63 2.35 2.85% 4.18 2.07 1.11% 2.16
Tesla 1.87 1.61% 1.70 3.51 2.55% 5.92 1.25 1.61% 2.08

The Comparative analysis on about LSTM (RNN), Facebook Prophet (Statistical Model) and Random Forest
Regressor (Ensemble Learning Model) and their performance in Table 3 shows the Recurrent Neural Network (LSTM)
provides better values for RMSE, MAPE and MAE.

CONCLUSIONS
This research explores the use of machine learning models, specifically LSTM, Facebook Prophet, and Random Forest
Regressor, to improve stock market sentiment forecasting. The study uses historical datasets from platforms like Yahoo
Finance and the company's website. LSTM is found to be a powerful tool for short-term forecasting, with impressive metrics
like RMSE, MAPE, and MAE. This suggests that LSTM holds promise for traders and investors in making informed
decisions in the fast-paced world of stock market investments. However, the study acknowledges its limitations due to the
unpredictable nature of financial markets and the ever-evolving landscape. The study contributes to the evolving discourse
on stock market prediction, providing a nuanced understanding of machine learning models' capabilities and limitations.
Further exploration and refinement of predictive techniques are essential for staying ahead in the dynamic financial market
landscape.

Author Contributions: Conceptualization, M.I.K.J.; Methodology, M.I.K.J.; Software, M.I.K.J.; Validation, M.I.K.J.; Formal Analysis, M.I.K.J.;
Investigation, M.I.K.J.; Resources, M.I.K.J.; Data Curation, M.I.K.J.; Writing – Original Draft Preparation, M.I.K.J.; Writing – Review & Editing,
M.I.K.J.; Visualization, M.I.K.J.; Supervision, M.I.K.J.; Project Administration, M.I.K.J.; Funding Acquisition, M.I.K.J. Authors have read and agreed to
the published version of the manuscript.
Institutional Review Board Statement: Ethical review and approval were waived for this study, due to that the research does not deal with vulnerable
groups or sensitive issues.
Funding: The authors received no direct funding for this research.
Acknowledgments: Throughout the duration of the course, I wish to extend my profoundest appreciation to my course supervisor, Dr. Md. Zahedul Hasan,
for his indispensable counsel, encouragement, and mentorship. I would like to express my sincere gratitude to Lecturer Aparna Das for her invaluable
5
 Jabed, American International Journal of Sciences and Engineering Research 7(1) (2024), 1-6

assistance in the domain of Machine Learning, which significantly advanced this research. Associate Professor K. M. Anwarul Islam a top-level scientist
who secured the 2nd position on the AD Scientific Index, World Scientist Rankings 2023 for Bangladesh in the field of Business & Management / Business
Administration at my university, deserves special recognition, for providing invaluable guidance, support, and assistance throughout the completion and
publication of this paper. He is also a Research Fellow and Visiting Professor at INTI School of Business, INTI International University, Malaysia. He has
authored or presented over 150+ scientific papers in reputed journals indexed in ABDC, ERA, ABS, Scopus-Q1/Q2, WoS-SCIE, and SSCI. His astute
advice and constructive feedback were significant in ensuring the publication success of this manuscript.
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.
Data Availability Statement: The data presented in this study are available on request from the corresponding author. The data are not publicly available
due to restrictions.
Conflicts of Interest: The authors declare no conflict of interest.

REFERENCES
Akhtar, M. M., Zamani, A. S., Khan, S., Shatat, A. S. A., Dilshad, S., & Samdani, F. (2022). Stock market prediction based
on statistical data using machine learning algorithms. Journal of King Saud University-Science, 34(4), 101940.
Dhaka Stock Exchange (DSE). (2023). Dhaka Stock Market Historical Data. Retrieved from https://www.dsebd.org
Hiransha, M., Gopalakrishnan, E. A., Menon, V. K., & Soman, K. P. (2018). NSE stock market prediction using deep-
learning models. Procedia computer science, 132, 1351-1362.
Pai, P. F., & Lin, C. S. (2005). A hybrid ARIMA and support vector machines model in stock price forecasting. Omega,
33(6), 497-505.
Yoon, Y., & Swales, G. (1991, January). Predicting stock price performance: A neural network approach. In Proceedings
of the twenty-fourth annual Hawaii international conference on system sciences (Vol. 4, pp. 156-162). IEEE.
World Bank (2018). Market capitalization of listed domestic companies (current US$). Retrieved from
https://data.worldbank.org/indicator/CM.MKT.LCAP.CD

Publisher’s Note: ACSE stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2024 by the authors. Licensee ACSE, USA. This article is an open access article distributed under the terms and conditions of the Creative Commons
Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

American International Journal of Sciences and Engineering Research (E-ISSN 2641-0311 P-ISSN 2641-0303) by ACSE is licensed under a Creative
Commons Attribution 4.0 International License.

View publication stats