Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Probabilistic LSTM Modeling for Stock Price

Prediction with Monte Carlo dropout Long
Short-Term Memory Network
Clement Asare*1, Derrick Asante2, John Fiifi Essel3
1
Department of Statistics & Actuarial Science, Kwame Nkrumah University of Science & Technology, Kumasi-Ghana
2
Department of Statistics, University of Toledo, Ohio-USA
3
Department of Mathematics & Statistics, Portland State University, Portland, OR - USA

Abstract:- Investors and financial professionals in today's feature extraction methods [8]. Deep learning algorithms have
dynamic stock markets attach great significance to precise emerged as attractive option among these methods since they
forecasts of stock returns. This emphasis is not only on are adaptable and produce a high accuracy level [28], [24],
accurate forecasting models but also their reliability. While [11], [22], [8], [2]. However, most machine learning
conventional machine learning models can predict algorithms rely on point estimates, providing a single
nonlinear datasets with high accuracy, they often overlook prediction without considering uncertainty. This can lead to a
uncertainties in their predictions, leading to unreliable lack of transparency and potentially misleading results. Many
outcomes. This study employed the Bayesian LSTM (Long researchers have attempted to develop an accurate stock price
Short-Term Memory) model for stock price prediction and forecasting method from linear to nonlinear models and
examined its performance with that of the conventional diverse machine learning algorithms. However, the concept of
LSTM model. The findings revealed that the Bayesian more resilient techniques continue to emerge. Utilizing the
LSTM model produces better results than the conventional Auto-regressive Integrated Moving Average (ARIMA) model
LSTM model considering the 𝑹𝟐 (R-squared), MAPE to forecast the prices of stocks on the New York Stock
(Mean Absolute Percentage Error), and RMSE (Root Exchange (NYSE) and National Stock Exchange (NSE), [1]
Mean Squared Error) values. The Bayesian LSTM model discovered promising results for short-term forecasting.
acknowledged the presence of inherent uncertainty in the Advanced machine learning and deep learning strategies for
underlying data and refrained from exhibiting excessive stock price forecasting were proposed by [27]. Their results
confidence in its predictions, having a confidence level of demonstrated that a deep learning LSTM network combined
48.67% in predicting the S&P 500 data for the study. This with high gradient boosting performs better than the
study provides a more reliable approach for stock price traditional ARIMA approach. [26] employed various machine
prediction to help investors and financial professionals learning techniques and conventional statistical methodologies
make informed decisions. to forecast Karachi Stock Exchange (KSE) market
performance. According to their findings, the multilayer
Keywords:- Bayesian LSTM Model; Monte Carlo Dropout; perceptron (MLP) algorithm outperforms the benchmark
Stock Price Prediction; LSTM Model; Machine Learning. methods. In a study by [17], different forecasting models were
compared, and it was discovered that the convolutional neural
I. INTRODUCTION network-Long Short-Term Memory (CNN-LSTM) approach
achieved the highest prediction accuracy for stock prices. [7]
Investors and financial professionals today place high addressed the challenge of stock price forecasting in dynamic
importance on accurate stock return forecasts in the ever- situations by proposing an LSTM-based model, which
changing world of stock markets. This results from the stock achieved good accuracy with low error rates when trained on a
prices' patterns being nonlinear and volatile [2]. One of the sizable dataset. In a recent study by [4], the LSTM model was
most crucial institutions in any economy is the stock market used to predict the next day closing price of the S&P 500
[8]. Therefore, it is essential to design a reliable model that index. Their findings indicate that a single-layer LSTM model
can capture the nonlinear nature and produce credible outperforms multilayer LSTM models, offering a better fit and
accuracy. To forecast stock prices, researchers have explored a higher prediction accuracy for the task. [20] also employed the
variety of statistical models [1], [12], [23], [19], but their transformer model to forecast the future stock prices on the
accuracy cannot be compared to machine learning techniques Dhaka Stock Exchange (DSE), the prominent stock exchange
[14], [21]. This is because machine learning can better capture in Bangladesh. The transformer algorithm exhibited strong
nonlinear patterns and produce accurate results [13]. Machine performance with promising accuracy, surpassing the ARIMA
learning techniques for stock prediction have recently attracted model's forecasting capabilities from the study's findings.
a lot of interest to improve their performance through various

IJISRT23JUL1502 www.ijisrt.com 2316

Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Although these machine learning techniques are flexible in financial time series data [18]. However, traditional LSTM
and have great potential in handling nonlinear data, the models lack uncertainty quantification in predictions. To
problem of uncertainty still needs to be considered. While overcome this limitation, the Bayesian LSTM model has
previous research has explored various machine learning and emerged as an effective approach, incorporating Bayesian
deep learning techniques [7], [27], [17], [20], we recognized inference to provide rigorous uncertainty estimates [10].
the significance of this factor in enhancing prediction accuracy Combining the strengths of LSTM and Bayesian inference, the
and minimizing biases. In this study, we propose a framework Bayesian LSTM model presents a robust framework for
that treats LSTM weights as random variables and estimates accurate stock price prediction with quantified uncertainty. In
posterior distributions by combining the strengths of LSTM this subsection, we establish the models of LSTM and
and Bayesian models in capturing sequential dependencies and Bayesian LSTM for stock price prediction, laying the
quantifying uncertainty, respectively. This will enable us to groundwork for subsequent analysis and evaluation of their
provide probabilistic forecasts in addition to point estimates, performance in forecasting financial markets.
aiding in model transparency and interpretability and fostering
trust in the forecasts of the LSTM to facilitate a more practical B. The Long Short Term Memory Model (LSTM).
way of communicating results. Based on the literature, LSTM model is an extension of recurrent neural network
research gaps exist in predicting the prices of stocks, where (RNN) that excels in capturing and modeling long-range
this study would make significant contributions. These gaps dependencies in sequential data, making it particularly suitable
include limited exploration of Bayesian approaches, the need for time series analysis [24], [18]. Unlike traditional RNNs,
for uncertainty estimation, handling dynamic situations with LSTM networks combine specialized memory cells with
fluctuating values, and comparison with other contemporary input, output, and forget gates, which allows them to maintain
models. By addressing these gaps, this paper would enhance relevant information over extended periods and selectively
the accuracy and interpretability of stock price predictions, update or forget information as needed. Mathematically, an
provide probabilistic forecasts, and offer insights into the input 𝑋 = (𝑥1 , 𝑥2 , … , 𝑥𝑛 ), where each 𝑥𝑡 , 𝑡 = 1, 2, 3, … , 𝑛 ∈
benefits of Bayesian inference in financial market forecasting. 𝑅𝑇 , 𝑛 is the number of input dimensions and 𝑇 represents the
This study contributes to the existing literature by introducing time lag. With an output 𝑦 = (𝑦1, 𝑦2 , … , 𝑦𝑛 ), the forward
uncertainty-aware machine learning algorithms to stock price training method of the LSTM model, as proposed by [16] can
predictions, delivering valuable insights for decision-making be formulated as follows:
in financial markets, and guiding future research. The
Bayesian LSTM model was compared with the regular LSTM 𝐼𝑡
model to measure the performance of both models in = σ(𝑊𝑡 ⋅ [𝑚𝑡−1 , 𝑥𝑡 ]
predicting stock prices. Other sections of the paper include + 𝑏𝑡 ) (1)
Section II, which describes our research methodology for both 𝐹𝑡
the regular LSTM and the Bayesian LSTM models. Section III = σ(𝑊𝑓 ⋅ [𝑚𝑡−1 , 𝑥𝑡 ]
presents our findings and a discussion; Section V concludes + 𝑏𝑓 ) (2)
the study.
𝐶𝑡
II. MATERIALS AND METHODS = 𝐹𝑡 ⋅ 𝐶𝑡−1 + 𝐼𝑡
⋅ tanh(𝑊𝑐 ⋅ [𝑚𝑡−1 , 𝑥𝑡 ]
Our data set contains daily historical prices of Standard + 𝑏𝑐 ) (3)
and Poor's 500 (S&P 500) from June 2017 to June 2022 which 𝑂𝑡
is publicly available on yahoo finance website. The data = σ(𝑊0 ⋅ [𝑚𝑡−1 , 𝑥𝑡 ]
include Open, High, Low, Close, Adjusted Close, and volume + 𝑏0 ) (4)
as features. To scale the values to be in the range [0, 1], we 𝑚𝑡
utilized the minimum-maximum scaler represented = 𝑂𝑡 ⋅ tanh(𝐶𝑡 ) (5)
mathematically as.
Where 𝐼𝑡 denotes the activation of the input gate, Ot
𝑋 − 𝑋𝑚𝑖𝑛 denotes the output gate, and Ft denotes the forget gate. Ct also
𝑋𝑆𝑐𝑎𝑙𝑒𝑑 = represents the activation vector for each cell, and mt
𝑋𝑚𝑎𝑥 − 𝑋𝑚𝑖𝑛
represents the memory block. The weight matrix and the bias
The data was split into training and testing sets: 80% for vector are defined as W and b, respectively. The hyperbolic
training and 20% for testing the performance of the models. tangent function and the sigmoid function represented by tanh
and σ respectively, are the two primary activation functions
A. The Model Establishment. the LSTM model considers. The equations for these two
Machine learning algorithms have become robust tools activation functions are represented as follows:
for stock price prediction recently. The Long Short-Term
Memory (LSTM) model, to be specific, is well known for 𝑒𝑥
σ(𝑥) =
incorporating temporary dependencies and non-linear patterns 𝑒𝑥 + 1

IJISRT23JUL1502 www.ijisrt.com 2317

Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

𝑒 2𝑥 − 1 Where xt and yt represents the input data at time t and its

tanh(𝑥) = corresponding output data at time t respectively. w ̂ is also
𝑒 2𝑥 + 1
sampled through dropout technology based on the Monte
C. The Bayesian Deep Learning Model (Bayesian-LSTM). Carlo integration.
The Bayesian LSTM model is an LSTM model which
incorporates Bayesian inference techniques. It utilizes D. Evaluation Metrics.
techniques such as Variational Inference, Gaussian Processes, The performance of LSTM model and the Bayesian-
and Monte Carlo Dropout to estimate uncertainty in LSTM model in predicting stock prices are measured using
predictions. In this paper, we used the Monte Carlo Dropout evaluation metrics such as Mean Absolute Percentage Error
proposed by [9] for the uncertainty estimation. During the (MAPE), Coefficient of Determination (R2 ), and Root Mean
training process, the Monte Carlo Dropout randomly Squared Error (RMSE) in this study. These metrics provide
deactivates some LSTM units (neurons) by setting them to quantitative measures of the models' accuracy and
zero, effectively excluding them from the network. This effectiveness in capturing the underlying patterns and trends in
procedure is iterated multiple times, generating distinct stock price data. By comparing the values of these metrics, we
dropout masks for each iteration. By training the model using can evaluate and compare the predictive performance of the
this dropout mechanism, the Bayesian LSTM can generate LSTM and Bayesian-LSTM models for stock price
predictions resilient to the absence of specific neurons, thereby forecasting. Mathematically, they are represented in Equations
improving its generalization capability and preventing 9-11. The closer the MAPE and RMSE to zero, the more
overfitting. By performing Monte Carlo simulations, the accurate the model is, and the closer the 𝑅2 to 1, the more
Bayesian LSTM generates a set of predictions, each associated precise the model is.
with a distinct set of sampled weights. The mean prediction is
the central estimate for the model's output and is calculated by 𝑀𝐴𝑃𝐸
𝑛
finding the mean of all the forecasts generated during the 1 𝑦𝑡 − 𝑦̂𝑡
simulations. It is expressed mathematically in Equation 8. = ∑| |
𝑛 𝑦𝑡
𝑡=1
Given an input 𝑋 = (𝑥1 , 𝑥2 , … , 𝑥𝑛 ), where each 𝑥𝑡 , 𝑡 = × 100 (9)
1, 2, 3, … , 𝑛 ∈ 𝑅 𝑇 , 𝑛 is the number of input dimensions and 𝑇
represents the time lag. With an output 𝑦 = (𝑦1 , 𝑦2, … , 𝑦𝑛 ), the 𝑅𝑀𝑆𝐸
Bayesian method aims to find the parameter 𝑤 of the function 𝑛
y = f w (x) which produces the optimal results [3], [15], [10]. 1 2
= √ ∑(𝑦𝑖 − 𝑦̂)
𝑖 (10)
Viewing the LSTM as a probability model, we then consider 𝑛
𝑖=1
𝑤 to be a random variable that follows the normal prior
distribution. The posterior probability distribution denoted by
𝑝(𝑤|𝑥, 𝑦) is based on the Bayesian principle. Calculating the 𝑅2
posterior distribution 𝑝(𝑤|𝑥, 𝑦) is challenging and quite ∑𝑛𝑡=1(𝑦𝑡 − 𝑦̂𝑡 )2
= 1− 𝑛 (11)
impossible to obtain directly [6], [15]. To address this, [6] ∑𝑡=1(𝑦𝑡 − 𝑦̅)2
proposed an alternative variational distribution 𝑞(𝑤|θ)
parameterized by θ that can effectively approximate the actual Where, 𝑛 represents the total number of observations or
posterior distribution derived from the available data. The data points, yt represents the actual value of the observation at
prediction results for the Bayesian LSTM, as proposed by [6], time 𝑡, ŷt represents the predicted value of the observation at
are obtained by employing stochastic propagation and Monte time 𝑡, and y̅ represents the mean of the actual values.
Carlo integration for the output mean and can be expressed as.
III. RESULTS AND DISCUSSION
𝑝(𝑦𝑡 |𝑥𝑡 , 𝑥, 𝑦)
= ∫ 𝑝(𝑦𝑡|𝑥𝑡 , 𝑤) A. Empirical Results.
This section presents the empirical findings derived from
⋅ 𝑝(𝑤|𝑥, 𝑦)𝑑𝑤 (6) our study on probabilistic stock price prediction. Our analysis
employed a Bayesian LSTM model to collect and analyze
𝑝(𝑦𝑡 |𝑥𝑡 , 𝑥, 𝑦) secondary data from the S&P 500. The result obtained from
≈ ∫ 𝑝(𝑦𝑡|𝑥𝑡 , 𝑤) this study offers valuable insights into the importance of
uncertainty quantification when predicting stock prices. In the
⋅ 𝑞(𝑤|θ)𝑑𝑤 (7) Bayesian LSTM model, the number of LSTM units was set to
𝑝(𝑦𝑡 |𝑥𝑡 , 𝑥, 𝑦) 50 to determine the hidden state's dimensionality. The dense
𝑁
1 layer's units were set to 1, aligning with the prediction of a
≈ ∑ 𝑝(𝑦𝑡 |𝑥𝑡 , 𝑤
̂) (8) single target variable (Close price). We used the Adam
𝑁
𝑡=1 optimizer to update the model weights during training, a

IJISRT23JUL1502 www.ijisrt.com 2318

Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
popular choice for deep-learning models. To determine the
number of samples taken from the Bayesian model to estimate
uncertainty, we set the number of Monte Carlo samples to 100.
These choices were made based on standard practices and
experimentation.

The Price dynamics of the data for this study (S&P 500)
are shown in Figure 1. Even though there are dynamics in the
price movement, there is evidence of an increasing trend, as
indicated by the Mann-Kendall test in Table 1.

Fig 1. The Close price Dynamics of S&P data

Table 1. Mann-Kendall Test.

FEATURE TREND S Var (S) Z p-value
Open increasing 654838 232743467.0 42.9 0.0
High increasing 661344 232743467.0 43.3 0.0
Low increasing 647001 232743467.0 42.4 0.0
Close increasing 654293 232743467.0 42.9 0.0
Adjusted Close increasing 654293 232743467.0 42.9 0.0
Volume increasing 306398 232743467.0 20.1 0.0

Table 2 shows the Bayesian LSTM model and predictions. Additionally, Figure 5 displays the forecast's
conventional LSTM models' performance metrics. From the posterior probability distribution, which indicates that the
table, the Bayesian LSTM model offers a better performance model assigns higher probability to the prices that fall between
with a lower MAPE (0.0097), RMSE (47.82), and a higher 𝑅2 $ 4,200 and $ 4,800. The broad tail on the left of the posterior
(0.9594) compared to the LSTM model with MAPE (0.0154), probability distribution graph suggests a larger span of
RMSE (80.00) and 𝑅2 (0.8865). These results indicate that the potential outcomes for lower stock prices. This indicates that
Bayesian LSTM model provides more accurate and reliable accurately forecasting these lower values poses more
predictions, capturing a higher percentage of the variance in difficulties for the model. This characteristic of the Bayesian
the target variable, which is also evident in the residual plot LSTM model proves advantageous, especially in scenarios
shown in Figure 4. involving significant fluctuations or volatility in the data. It
enables decision-makers to consider various potential
Table 2 Performance Metrics outcomes, enhancing their ability to make informed choices.
MODEL MAPE 𝑹𝟐 RMSE
LSTM 0.0154 0.8865 80.00
BAYESIAN
0.0097 0.9594 47.82
LSTM

The Actual and the predicted plot of the two models are
shown in Figure 2, indicating the model's performance in
predicting the S&P 500. In Figure 2, both models seem to
capture the dynamic nature of the S&P 500 prices. However,
the Bayesian LSTM model captured the nonlinear nature of
the S&P 500 data for the study period's test set. With the
Bayesian LSTM model demonstrating high prediction
accuracy, the uncertainty inclusion provides valuable
information, as depicted in Figure 3. The figure illustrates how
the model acknowledges the presence of inherent uncertainty
in the underlying data and avoids excessive confidence in its

IJISRT23JUL1502 www.ijisrt.com 2319

Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig 4. Residual plots of the two Models

Fig 2. The two models' Actual and Predicted graphs for the
test set

Fig 3. The Bayesian LSTM prediction with Uncertainty. Fig 5. Posterior Probability distribution of the Prediction with
uncertainty

B. Discussion.
The study's outcomes suggest that the proposed
probabilistic LSTM model incorporating Monte Carlo dropout
demonstrates promising performance in forecasting stock
prices. By introducing uncertainty through dropout during
training, the model becomes better suited to address the
inherent unpredictability of financial markets. The model's
capability to provide probabilistic predictions is particularly
valuable for decision-makers, enabling them to grasp the range
of potential outcomes and make well-informed choices. These
findings align with existing literature such as [6], [5], [10], and
[3] on the advantages of probabilistic modeling for prediction
compared to traditional deterministic methods. The adoption
of Monte Carlo dropout in LSTM networks has been effective
in capturing uncertainty, consistent with other studies that

IJISRT23JUL1502 www.ijisrt.com 2320

Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
have utilized dropout-based techniques for uncertainty REFERENCES
estimation in deep learning models, such as [6].
[1]. Adebiyi A Ariyo, Adewumi O Adewumi, and Charles K
The ability to generate confidence intervals and Ayo. Stock price prediction using the arima model. In
probability distributions for predictions helps investors and 2014 UKSim- AMSS 16th international conference on
financial analysts to gain a comprehensive understanding of computer modelling and simulation, pages 106–112.
the risks associated with various investment decisions. While IEEE, 2014.
the results of this study are promising, it is crucial to [2]. Gourav Bathla. Stock price prediction using lstm and svr.
acknowledge the limitations and potential challenges of the In 2020 Sixth International Conference on Parallel,
Bayesian LSTM model. The model's performance may rely on Distributed and Grid Computing (PDGC), pages 211–
the quality and size of the training dataset, as well as the 214. IEEE, 2020.
selection of hyperparameters. Additionally, future research [3]. Jose´ M Bernardo and Adrian FM Smith. Bayesian theory,
should consider the uncertainties that arise from datasets volume 405. John Wiley & Sons, 2009.
(Aleatoric uncertainty) and uncertainties that arise from the [4]. Hum Nath Bhandari, Binod Rimal, Nawa Raj Pokhrel,
models (Epistemic uncertainty) as they can provide valuable Ram- chandra Rimal, Keshab R Dahal, and Rajendra KC
insight to help provide a more reliable prediction. By building Khatri. Predicting stock market index using lstm.
on the existing literature and addressing the uncertainties Machine Learning with Applications, 9:100320, 2022.
associated with stock price prediction, this study contributes to [5]. Jose Caceres, Danilo Gonzalez, Taotao Zhou, and
the advancement of the field and opens new avenues for future Enrique Lopez Droguett. A probabilistic bayesian
research in financial forecasting and deep learning recurrent neu- ral network for remaining useful life
applications. prognostics considering epistemic and aleatory
uncertainties. Structural Control and Health Monitoring,
IV. CONCLUSION 28(10), 2021.
[6]. Junfu Chen, Dechang Pi, Zhiyuan Wu, Xiaodong Zhao,
Accurate predictive models are valuable; however, their Yue Pan, and Qiang Zhang. Imbalanced satellite
dependability is paramount for investors and decision-makers. telemetry data anomaly detection model based on
Although conventional machine learning models exhibit bayesian lstm. Acta Astronautica, 180:232–242, 2021.
relatively high accuracy in predicting non-linear datasets, they [7]. Mohammad Diqi. Stocktm: Accurate stock price
often neglect uncertainties, resulting in unreliable outcomes. prediction model using lstm. International Journal of
This study explored and compared the Bayesian LSTM model Informatics and Computation, 4(1):1–10, 2022.
and conventional LSTM models' performance for stock price [8]. Jithin Eapen, Doina Bein, and Abhishek Verma. Novel
prediction. The results indicated the Bayesian LSTM is better deep learning model with cnn and bi-directional lstm for
than the conventional LSTM model using the MAPE (Mean improved stock market index prediction. In 2019 IEEE
Absolute Percentage Error) and 𝑅2 values as evaluation 9th annual computing and communication workshop and
metrics. The Bayesian LSTM model acknowledges the conference (CCWC), pages 0264–0270. IEEE, 2019.
presence of inherent uncertainty in the underlying data and [9]. Yarin Gal and Zoubin Ghahramani. Dropout as a
refrains from exhibiting excessive confidence in its bayesian approximation: Representing model uncertainty
predictions, it uses a confidence level of 48.67% for in deep learn- ing. In international conference on
predicting the S&P 500 data in this study. Every model has machine learning, pages 1050–1059. PMLR, 2016.
room for improvement, and we will strive to enhance the [10]. Fatemeh Ghobadi and Doosun Kang. Multi-step ahead
accuracy of our forecast results in future endeavors. probabilistic forecasting of daily streamflow using
bayesian deep learning: A multiple case study. Water,
DECLARATIONS 14(22):3672, 2022.
[11]. Achyut Ghosh, Soumik Bose, Giridhar Maji, Narayan
Funding Deb- nath, and Soumya Sen. Stock price prediction using
The authors received no specific funding for this work. lstm on indian share market. In Proceedings of 32nd
international conference on, volume 63, pages 101–110,
Declaration of Competing Interest 2019.
The authors have declared that no competing interests [12]. Banhi Guha and Gautam Bandyopadhyay. Gold price
exist. forecasting using arima model. Journal of Advanced
Management Science, 4(2), 2016.

IJISRT23JUL1502 www.ijisrt.com 2321

Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[13]. Samuel Asante Gyamerah, Philip Ngare, and Dennis [24]. Sreelekshmy Selvin, R Vinayakumar, EA
Ikpe. Probabilistic forecasting of crop yields via quantile Gopalakrishnan, Vijay Krishna Menon, and KP Soman.
random forest and epanechnikov kernel function. Stock price prediction using lstm, rnn and cnn-sliding
Agricultural and Forest Meteorology, 280:107808, 2020. window model. In 2017 international conference on
[14]. C Karthikeyan, P Anandan, R Prabha, D Mohan, advances in computing, commu- nications and
Vijendra Babu, et al. Predicting stock prices using informatics (icacci), pages 1643–1647. IEEE, 2017.
machine learning techniques. In 2021 6th International [25]. Xuanyi Song, Yuetian Liu, Liang Xue, Jun Wang,
Conference on Inventive Computation Technologies Jingzhe Zhang, Junqiang Wang, Long Jiang, and Ziyan
(ICICT), pages 1–5. IEEE, 2021. Cheng. Time- series well performance prediction based
[15]. Gaoyang Li, Li Yang, Chi-Guhn Lee, Xiaohua Wang, and on long short-term memory (lstm) neural network model.
Mingzhe Rong. A bayesian deep learning rul framework Journal of Petroleum Science and Engineering,
integrating epistemic and aleatoric uncertainties. IEEE 186:106682, 2020.
Trans- actions on Industrial Electronics, 68(9):8829– [26]. Mehak Usmani, Syed Hasan Adil, Kamran Raza, and
8841, 2020. Syed Saad Azhar Ali. Stock market prediction using
[16]. Xiang Li, Ling Peng, Xiaojing Yao, Shaolong Cui, Yuan machine learning techniques. In 2016 3rd international
Hu, Chengzeng You, and Tianhe Chi. Long short-term conference on computer and information sciences
memory neural network for air pollutant concentration (ICCOINS), pages 322–327. IEEE, 2016.
predictions: Method development and evaluation. [27]. Pham Hoang Vuong, Trinh Tan Dat, Tieu Khoi Mai,
Environmental pollution, 231:997–1004, 2017. Pham Hoang Uyen, et al. Stock-price forecasting based
[17]. Wenjie Lu, Jiazheng Li, Yifan Li, Aijun Sun, and on xgboost and lstm. Computer Systems Science &
Jingyang Wang. A cnn-lstm-based model to forecast Engineering, 40(1), 2022.
stock prices. Complexity, 2020:1–10, 2020. [28]. Ren Zi, Yin Jun, Yu Yicheng, Ma Fuxiang, and Li
[18]. Md Abdullah Al Mehedi, Marzieh Khosravi, Munshi Rongbin. Stock price prediction based on optimized
Md Shafwat Yazdan, and Hanieh Shabanian. Exploring random forest model. 2022 Asia Conference on
temporal dynamics of river discharge using univariate Algorithms, Computing and Machine Learning
long short-term memory (lstm) recurrent neural network (CACML), 2022.
at east branch of delaware river. Hydrology, 9(11):202,
2022.
[19]. Prapanna Mondal, Labani Shit, and Saptarsi Goswami.
Study of effectiveness of time series modeling (arima) in
forecasting stock prices. International Journal of
Computer Science, Engineering and Applications,
4(2):13, 2014.
[20]. Tashreef Muhammad, Anika Bintee Aftab, Muhammad
Ibrahim, Md Mainul Ahsan, Maishameem Meherin
Muhu, Shahidul Islam Khan, and Mohammad Shafiul
Alam. Transformer-based deep learning model for stock
price predic- tion: A case study on bangladesh stock
market. International Journal of Computational
Intelligence and Applications, page 2350013, 2023.
[21]. Jigar Patel, Sahil Shah, Priyank Thakkar, and Ketan
Kotecha. Predicting stock and stock price index
movement using trend deterministic data preparation and
machine learning tech- niques. Expert systems with
applications, 42(1):259–268, 2015.
[22]. Kriti Pawar, Raj Srujan Jalem, and Vivek Tiwari. Stock
market price prediction using lstm rnn. In Emerging
Trends in Expert Applications and Security: Proceedings
of ICETEAS 2018, pages 493–503. Springer, 2019.
[23]. Sanjiban Sekhar Roy, Dishant Mittal, Avik Basu, and
Ajith Abraham. Stock market forecasting using lasso
linear re- gression model. In Afro-European Conference
for Industrial Advancement: Proceedings of the First
International Afro- European Conference for Industrial
Advancement AECIA 2014, pages 371–381. Springer,
2015.

IJISRT23JUL1502 www.ijisrt.com 2322