Stock Prediction Using Machine Learning: Enhancing Market

Forecasting
Abstract

Stock market prediction remains a challenging task due to market volatility, investor sentiment,
and macroeconomic influences. Traditional forecasting methods often struggle to capture the
nonlinear and dynamic nature of stock prices. Machine learning techniques, particularly Long
Short-Term Memory (LSTM) networks, have demonstrated significant potential in improving
predictive accuracy by identifying patterns in historical data. This paper reviews the progress made
in developing an LSTM-based stock prediction model, which currently relies on a predefined
dataset. The next phase of development involves integrating real-time financial data from Yahoo
Finance to enable broader and more adaptive forecasting. While the model shows promising results,
challenges such as hyperparameter tuning, data preprocessing, and external market influences
remain. Future work will focus on optimizing model performance, incorporating sentiment
analysis, and enhancing computational efficiency to create a more robust and scalable prediction
framework.

Keywords: Stock market prediction, machine learning, LSTM, Yahoo Finance, financial
forecasting, time-series analysis, investor sentiment, deep learning, market volatility, real-time
data integration.

1. Introduction

Stock market prediction has been a longstanding challenge due to the highly dynamic and volatile
nature of financial markets. Traditional forecasting methods, such as fundamental and technical
analysis, offer valuable insights but often fail to capture the complex and nonlinear dependencies
present in stock price movements. With the advancements in machine learning (ML) and deep
learning, particularly Long Short-Term Memory (LSTM) networks, more sophisticated models
have been developed to improve stock price prediction accuracy by leveraging historical data and
identifying hidden patterns.

The purpose of this study is to build an LSTM-based stock market prediction model that utilizes
historical stock data for forecasting. The current implementation operates on a predefined dataset,
focusing on specific stocks with structured historical data. This dataset includes key financial
indicators such as stock opening and closing prices, trading volume, and adjusted closing values.
By leveraging the sequential nature of LSTMs, the model is capable of capturing long-term
dependencies in stock price trends, making it a valuable tool for investors seeking data-driven
decision-making support. While this approach provides valuable insights, its predictive scope
remains limited. [1][2]

To enhance real-world applicability, the next stage involves integrating Yahoo Finance to enable
real-time data retrieval and dynamic forecasting across multiple stocks. This integration will allow
the model to continuously update with the latest market data, making it more responsive to recent
price fluctuations and macroeconomic events. The combination of deep learning techniques with
real-time financial data is expected to significantly improve model performance and adaptability.
In addition to real-time data integration, further model refinements will involve hyperparameter
optimization, data preprocessing enhancements, and feature engineering to incorporate external
factors such as economic indicators, investor sentiment, and geopolitical events. These factors play
a crucial role in stock price movements and, when incorporated into the model, could improve
predictive accuracy and robustness.[1]

Moreover, stock market fluctuations are influenced by numerous external factors such as
macroeconomic conditions, global financial news, and investor sentiment, making stock price
prediction a complex problem. While LSTMs excel in capturing sequential dependencies,
additional methodologies, such as hybrid models incorporating sentiment analysis, attention
mechanisms, and reinforcement learning, may further enhance predictive accuracy.

This paper reviews the progress made in developing the stock market prediction model and
discusses key challenges, including data preprocessing, model optimization, and computational
efficiency. Future improvements will focus on refining model performance, improving real-time
data integration, and incorporating external factors for a more comprehensive forecasting
framework. The findings and future directions outlined in this study contribute to the growing field
of AI-driven financial analytics, aiming to provide valuable tools for investors, analysts, and
financial institutions.[2][3]

4. Methodology

The methodology for this study focuses on developing a stock market prediction model tailored
specifically for Ambuja Cement. The approach includes data collection, preprocessing, model
architecture, training, evaluation, and real-time implementation. Each stage is crucial in ensuring
the model accurately captures stock price patterns and industry-specific factors influencing
Ambuja Cement’s valuation. The cement industry operates within a complex framework
influenced by macroeconomic conditions, raw material availability, seasonal demand fluctuations,
and policy regulations. These variables are incorporated into the machine learning framework to
create a more precise and adaptable predictive model. Unlike general stock forecasting models that
rely solely on historical stock prices, this research incorporates operational and market-based
factors that significantly impact Ambuja Cement’s financial performance. By leveraging advanced
deep learning techniques, the model is designed to analyze past trends, predict stock movements,
and enhance investment decision-making for stakeholders.

4.1 Data Collection

The first step in developing the predictive model involves collecting extensive financial and
operational data related to Ambuja Cement. Stock market data, including daily opening and closing
prices, highest and lowest trading values, volume, and adjusted closing prices, is sourced from
Yahoo Finance to establish historical stock movement patterns. Given that cement stocks are
deeply influenced by broader economic indicators, the dataset is further enriched with
macroeconomic variables such as GDP growth, inflation rates, interest rate fluctuations, and
government infrastructure spending. Since Ambuja Cement’s revenue generation is closely tied to
the construction and infrastructure sectors, data on real estate growth, cement price indices, and
production capacity utilization are integrated to provide deeper insights into the company’s market
positioning.

Furthermore, cost drivers such as raw material prices—including limestone, gypsum, and fly ash—
along with fuel expenses, particularly coal and petcoke, are considered. Since energy accounts for
a significant portion of cement production costs, fluctuations in global crude oil prices,
transportation costs, and power tariffs are also incorporated into the dataset. Ambuja Cement’s
annual and quarterly financial reports are analyzed to extract key performance indicators such as
revenue, net profit margins, EBITDA, and debt-to-equity ratios. Additionally, sentiment analysis
is conducted on financial news, investor reports, and social media discussions to understand how
market perception and investor sentiment impact stock movements. The integration of these
multiple data sources ensures that the predictive model captures both fundamental and technical
factors influencing Ambuja Cement’s stock prices.
4.2 Data Preprocessing

Once the data is collected, extensive preprocessing is conducted to ensure it is clean, structured,
and optimized for machine learning. Stock market data often contains noise, missing values, and
inconsistencies due to market holidays, trading suspensions, and reporting discrepancies. Handling
missing values is a crucial step in ensuring data reliability. Missing price data is filled using
interpolation techniques, such as linear and polynomial interpolation, to maintain temporal
continuity. In cases where essential financial indicators are absent, mean imputation or regression-
based estimation methods are applied to fill the gaps. Additionally, outlier detection is performed
using statistical techniques like the interquartile range (IQR) method to identify and remove
extreme values that could distort model predictions.

Normalization is applied to bring all features to a common scale, ensuring that no single variable
dominates the learning process. Since stock prices and macroeconomic indicators operate on
different scales, Min-Max scaling is implemented to transform values into a fixed range between
0 and 1. This transformation preserves relative differences while preventing extreme fluctuations
from skewing predictions. Moreover, feature selection is performed to identify the most relevant
variables influencing Ambuja Cement’s stock prices. Techniques such as Pearson correlation
analysis, mutual information scoring, and Principal Component Analysis (PCA) are used to
eliminate redundant or less significant features. The resulting dataset is then structured into time-
series format, where past observations serve as input features for predicting future stock prices.
Given that cement industry trends often follow seasonal cycles, rolling window techniques are
employed to capture long-term dependencies and cyclical fluctuations. By implementing these
preprocessing steps, the dataset becomes more structured and optimized, allowing the model to
learn meaningful patterns and generate more accurate predictions.

4.3 Model Architecture

The predictive model is designed using a deep learning framework, specifically Long Short-Term
Memory (LSTM) networks, which have demonstrated high efficiency in time-series forecasting.
LSTM networks are a specialized form of recurrent neural networks (RNNs) that excel at capturing
long-term dependencies and sequential relationships in financial data. The model architecture
consists of multiple layers, each playing a distinct role in processing stock market information and
generating forecasts.

The input layer is responsible for processing historical stock prices, technical indicators, and
macroeconomic variables. This data is formatted into three-dimensional tensors with a shape
corresponding to batch size, time steps, and feature count, ensuring compatibility with LSTM
layers. Multiple stacked LSTM layers form the core of the model, allowing it to detect both short-
term fluctuations and long-term trends in Ambuja Cement’s stock movements. Each LSTM layer
retains important historical information while filtering out irrelevant noise, enabling the model to
recognize recurring patterns and predict future price trends more accurately. To prevent overfitting,
dropout layers are incorporated into the architecture, where a fraction of neurons is randomly
deactivated during training, promoting generalization and reducing the risk of memorizing training
data.

The final stage involves a fully connected dense layer, which aggregates the learned
representations and maps them to stock price predictions. A linear activation function is applied in
the output layer since stock prediction is a regression task that requires continuous numerical
outputs. Optimization techniques such as Adam optimizer are utilized to fine-tune model
parameters, ensuring faster convergence and improved accuracy. Additionally, hyperparameter
tuning is conducted using Bayesian optimization to determine the optimal number of LSTM units,
dropout rates, and learning rates. By refining these architectural components, the model achieves
a balance between complexity and predictive power, making it highly effective for forecasting
Ambuja Cement’s stock prices.

Hyperparameter Value Description

Number of LSTM
2 Multi-layer LSTM architecture
Layers
Number of Neurons
100 Number of LSTM units in each layer
per Layer
Activation Function Tanh Non-linear activation used in LSTM
Optimizer Adam Adaptive learning rate optimization
Mean Squared Error Measures the difference between actual and
Loss Function
(MSE) predicted values
Batch Size 32 Number of samples per training batch
Time Steps 60 Number of past days used for prediction

Table 1: LSTM Model Configuration and Hyperparameters

4.4 Model Training and Evaluation

The training phase involves feeding historical stock data into the LSTM network, allowing it to
learn patterns and relationships that drive Ambuja Cement’s stock price movements. The dataset
is split into training and testing subsets, typically in an 80:20 ratio, ensuring that the model
generalizes well to unseen data. Training is conducted using backpropagation, where errors in
predictions are propagated backward through the network to update weight parameters iteratively.
The model’s learning process is guided by the Mean Squared Error (MSE) loss function, which
measures the average squared difference between actual and predicted prices, ensuring that large
deviations are penalized more heavily.

To prevent overfitting and enhance generalization, early stopping is employed, where training halts
if validation loss stops improving, preventing unnecessary computations and reducing the risk of
model memorization. Evaluation metrics such as Root Mean Squared Error (RMSE), Mean
Absolute Error (MAE), and R-Squared scores are used to assess the model’s predictive accuracy.
A comparative analysis is conducted to benchmark the LSTM model against traditional forecasting
methods such as Auto-Regressive Integrated Moving Average (ARIMA) and Exponential
Smoothing. Results indicate that the LSTM-based approach outperforms conventional statistical
models by effectively capturing complex dependencies and nonlinear trends. However,
performance variations are observed during high-volatility periods, highlighting the need for
further refinements in feature engineering and external factor integration.
 Fig 1: Model Training and Validation Loss Curve

4.5 Real-Time Implementation

To enhance practical usability, the predictive model is deployed as a real-time forecasting tool
accessible through a web-based interface. The backend is developed using Flask, handling requests,
retrieving live stock data, and processing it through the trained LSTM model. The frontend, built
with HTML, CSS, and JavaScript, provides an interactive platform where users can input stock
symbols and receive real-time predictions. The system is designed for scalability, allowing
integration with additional financial indicators and investor sentiment analysis for improved
forecasting capabilities. By continuously updating with the latest stock market data, the model
adapts to changing market conditions, making it a valuable tool for investors, financial analysts,
and industry stakeholders seeking data-driven decision-making support

5. Results and Discussion

The predictive capability of the model has been assessed through various performance metrics,
including Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and R-Squared scores,
to determine its accuracy and reliability in forecasting stock prices. The initial evaluation indicates
that the LSTM-based model consistently surpasses traditional forecasting techniques such as Auto-
Regressive Integrated Moving Average (ARIMA) and simple Moving Averages in capturing the
intricate, long-term dependencies inherent in stock market behavior. Unlike statistical models,
which rely on linear relationships and struggle with nonlinear patterns, LSTM networks excel in
identifying complex sequences and interdependencies, making them particularly effective for
stock market prediction. The results demonstrate a strong correlation between predicted and actual
stock prices, especially in stable market conditions, where deviations are minimal, and predictions
remain aligned with historical trends. However, the model's accuracy tends to decline during
periods of heightened volatility, signaling the need for further refinements to improve its
adaptability to rapid and unpredictable market movements. Additionally, the model demonstrates
robust performance when applied to stocks with well-established historical trends, providing
reliable forecasts for companies with stable financial structures and consistent revenue growth.
However, its predictive strength diminishes when analyzing speculative or highly volatile stocks,
suggesting that adjustments in hyperparameters, feature selection, and model architecture could
further optimize results. Comparative experimentation with varying configurations indicates that
increasing the number of LSTM layers and adjusting dropout rates may contribute to enhanced
performance, improving the model’s ability to process intricate dependencies and refine
predictions.

Fig 2: Current Output of the Model

Challenges Faced

Despite its promising performance, the model encounters several challenges that impact its overall
effectiveness. One of the primary issues is overfitting, where the model performs exceptionally
well on training data but struggles to generalize effectively on unseen test data. Regularization
techniques such as dropout layers and L2 weight penalties have been implemented to address this,
but fine-tuning these techniques remains a complex task requiring extensive experimentation.
Hyperparameter tuning is another significant challenge, as selecting the optimal combination of
learning rate, batch size, and the number of LSTM units necessitates iterative testing to achieve
the best predictive performance.

Computational cost is another limitation, as training deep learning models on large datasets
requires substantial processing power and time. While GPU acceleration has improved efficiency,
further optimization is needed to enhance real-time prediction capabilities. Additionally, ensuring
the model’s adaptability across different industries remains a challenge, as some sectors exhibit
greater volatility than others. Future improvements will focus on refining feature engineering by
incorporating industry-specific financial indicators and external macroeconomic variables to
improve the model’s predictive accuracy and generalizability.

Interpretation of Trends

The model effectively captures stock market trends by analyzing historical data and recognizing
recurring patterns. It identifies support and resistance levels, moving averages, and seasonal
variations in stock prices. The ability of LSTMs to remember long-term dependencies enhances
the model’s predictive power, particularly in recognizing momentum shifts. However, while the
model accurately forecasts short-term price movements, it struggles with unexpected market
shocks caused by geopolitical events, economic news, or sudden changes in investor sentiment.
These limitations suggest that incorporating external indicators, such as sentiment analysis from
financial news and macroeconomic data, could improve robustness. Moreover, testing the model’s
effectiveness on different stock market indices has shown that its performance varies based on
market liquidity and trading volume. Enhancing the model’s adaptability to different market
conditions by dynamically adjusting learning rates or incorporating reinforcement learning
strategies could lead to more consistent predictions.

Limitations of Current Results

While the model yields promising predictive outcomes, it remains constrained by several inherent
limitations that require further exploration and enhancement. A key limitation lies in its reliance
on historical stock prices as the primary determinant of future price movements. Although past
trends provide valuable insights into market behavior, stock markets are highly dynamic, and
future fluctuations are often influenced by external variables that historical data alone cannot
predict. This reliance on past price patterns restricts the model’s ability to account for economic
disruptions, global financial crises, or sudden shifts in investor sentiment, limiting its effectiveness
in highly volatile markets. Additionally, slight deviations in predictive accuracy during periods of
increased market uncertainty suggest that the model requires further refinement to improve its
stability and robustness in dynamic trading environments.

Another limitation involves the assumption that historical price trends serve as a reliable indicator
of future stock performance. While LSTMs excel at identifying sequential dependencies, financial
markets are influenced by a wide array of external factors, including political events, central bank
policies, currency fluctuations, and sector-specific developments. Addressing this limitation
requires the integration of supplementary data sources, such as macroeconomic indicators,
financial sentiment analysis, and news-based market insights, to provide a more comprehensive
forecasting framework. Furthermore, testing the model across multiple industries has revealed that
certain stocks respond more predictably to LSTM-based forecasting, while others, particularly
those characterized by speculative trading, exhibit lower predictive reliability. This discrepancy
underscores the need for adaptive modeling strategies that account for industry-specific risk factors
and external influences
 Fig 3: Confusion Matrix for Stock Price Movement Classification

6. Future Work & Model Completion Strategy

Integration of Yahoo Finance API

To enhance the model’s capability for real-time forecasting, the integration of the Yahoo Finance
API is a critical next step. This will allow the system to fetch live stock market data, ensuring that
predictions are based on the most recent market conditions rather than static historical datasets.
Real-time data retrieval will improve the model’s responsiveness to sudden market shifts,
economic news, and global events that impact stock prices. Additionally, the API integration will
enable the automation of data updates, reducing manual intervention and enhancing forecasting
efficiency.

Feature Engineering Improvements

Further enhancements in feature engineering will be implemented to increase prediction accuracy.

The inclusion of investor sentiment analysis derived from financial news and social media
platforms will provide insights into market perception and potential volatility. Macroeconomic
indicators such as GDP growth, inflation rates, and central bank policies will also be incorporated
into the model, as these factors significantly impact stock price movements. By enriching the
dataset with diverse financial indicators, the model can better capture external influences and
improve long-term prediction stability. Furthermore, additional technical indicators such as
moving average crossovers, momentum oscillators, and volatility indices will be explored to refine
trend detection and improve decision-making capabilities.
Optimization Strategies

To further improve model performance and reduce computational costs, advanced optimization
strategies will be employed. Bayesian optimization will be utilized for hyperparameter tuning,
ensuring that the model selects the most effective configurations for learning rate, batch size, and
the number of LSTM layers. Transfer learning techniques will also be explored, leveraging pre-
trained models on financial datasets to enhance predictive performance without extensive
retraining. Additionally, reinforcement learning may be integrated, allowing the model to adapt
dynamically to changing market conditions by continuously updating prediction strategies based
on real-time feedback.

Strategy Description Expected Benefit

Bayesian Optimization Automated hyperparameter Improves accuracy while

tuning to find the best model reducing trial-and-error tuning
settings
Transfer Learning Using pretrained models on Reduces training time and
financial dataset to fine tune
enhances predictive
performance
Reinforcement Learning Adapting model strategies Increases model adaptability
dynamically based on market to changing trends
conditions
Gradient Clipping Prevents exploding gradients Stabilizes model training and
during backpropagation prevents sudden weight shifts
Early Stopping Stops training when validation Avoids overfitting and
loss stops improving reduces computation costs

Table 2: Optimization Strategies for Enhancing Model Performance

Deployment Plans

The final phase of model development includes deploying the stock prediction system as a web-
based or mobile application. The web-based platform will be built using Flask and Django for
backend processing, with a user-friendly front end developed in React or Vue.js. This will enable
users to input stock symbols and receive instant predictions, along with visualized historical trends
and future forecasts. A mobile application will also be considered, providing investors and traders
with on-the-go access to stock market predictions. Cloud deployment on platforms such as AWS
or Google Cloud will ensure scalability and efficiency, allowing the model to handle large user
requests while maintaining low latency. Security measures, including encrypted data transmission
and user authentication, will be implemented to safeguard financial data and ensure reliable access
for users.

By implementing these enhancements, the stock market prediction model will become more robust,
adaptable, and user-friendly, providing valuable insights to investors and financial analysts.
7. Conclusion

This research has demonstrated the effectiveness of Long Short-Term Memory (LSTM) networks
in predicting stock prices by leveraging historical financial data and advanced deep learning
techniques. The model successfully captures temporal dependencies and nonlinear patterns that
traditional forecasting methods, such as ARIMA and Moving Averages, often fail to recognize.
Through rigorous evaluation, the model has shown strong predictive accuracy, particularly for
stocks with well-defined historical trends. However, challenges remain in handling market
volatility, hyperparameter tuning, and incorporating external factors such as economic events and
investor sentiment.

One of the model’s key strengths lies in its ability to process sequential stock data efficiently and
extract meaningful insights. The use of multiple LSTM layers ensures that both short-term
fluctuations and long-term trends are captured, leading to improved forecasting capabilities.
Additionally, the integration of dropout layers and regularization techniques helps mitigate
overfitting, ensuring that the model generalizes well to unseen data. The inclusion of various
evaluation metrics, such as Mean Squared Error (MSE) and Root Mean Squared Error (RMSE),
provides a comprehensive performance assessment, validating the model’s effectiveness against
traditional statistical approaches. Moreover, the planned integration of real-time data sources via
the Yahoo Finance API will further enhance the model’s predictive power by allowing dynamic
adjustments to changing market conditions.

Despite its strengths, the model requires several refinements for full implementation. The first
major step is the seamless integration of external financial indicators, including macroeconomic
variables, investor sentiment analysis, and news-based market insights. These additional features
will help the model account for sudden market shifts that purely historical data cannot predict.
Secondly, further optimization strategies, such as Bayesian optimization and reinforcement
learning, will be applied to enhance model adaptability and efficiency. Computational performance
improvements, including optimizing model execution speed and reducing resource requirements
through cloud deployment, are also necessary to ensure scalability. Finally, deploying the model
as a fully functional web and mobile application will provide end-users with a user-friendly
interface to access real-time stock predictions.

This research underscores the growing role of machine learning in financial forecasting, offering
a powerful tool for investors, analysts, and financial institutions seeking data-driven decision-
making solutions. By combining LSTM networks with real-time data integration and advanced
optimization techniques, the model presents a robust framework for stock market prediction.
Although challenges such as market unpredictability and computational complexity persist,
ongoing enhancements will continue to refine the model’s performance and practical usability.
Ultimately, this research lays the groundwork for future advancements in AI-driven financial
analytics, bridging the gap between deep learning methodologies and real-world market
applications.
8. References

[1] M. Vijh, D. Chandola, V. A. Tikkiwal, and A. Kumar, "Stock Closing Price Prediction using
Machine Learning Techniques," Procedia Computer Science, vol. 167, pp. 599–606,2020, doi:
10.1016/j.procs.2020.03.326.

[2] M. M. Kumbure, C. Lohrmann, P. Luukka, and J. Porras, "Machine learning techniques and
data for stock market forecasting: A literature review," Expert Systems With Applications, vol.
197, p. 116659, 2022, doi: 10.1016/j.eswa.2022.116659.

[3] J. Shen and M. O. Shafiq, "Short-term stock market price trend prediction using a
comprehensive Machine Learning system," J. Big Data, vol. 7, no. 66, pp. 1-33, 2020,
doi:10.1186/s40537-020-00333-6.

[4] M. Nabipour, P. Nayyeri, H. Jabani, S. Shamshirband, and A. Mosavi, "PredictingStock Market

Trends Using Machine Learning and Machine Learning Algorithms Via Continuous and Binary
Data; a Comparative Analysis," IEEE Access, vol. 8, pp.150199-150212, 2020, doi:
10.1109/ACCESS.2020.3015966.

[5] A. Moghar and M. Hamiche, "Stock Market Prediction Using LSTM Recurrent
NeuralNetwork," Procedia Computer Science, vol. 170, pp. 1168-1173, 2020.
doi:10.1016/j.procs.2020.03.049.

[6] M. Leippold, Q. Wang, and W. Zhou, "Machine Learning in the Chinese Stock Market,"
Journal of Financial Economics, vol. 145, no. 1, pp. 64-82, 2022. Doi:
10.1016/j.jfineco.2021.08.017.

[7] G. Ding and L. Qin, "Study on the prediction of stock price based on the associated network
model of LSTM," International Journal of Machine Learning and Cybernetics, vol. 11 pp. 1307-
1317, 2020, doi: 10.1007/s13042-019-01041-1.

[8] W. Lu, J. Li, Y. Li, A. Sun, and J. Wang, "A CNN-LSTM-Based Model to Forecast Stock
Prices," Complexity, vol. 2020, Article ID 6622927, 10 pages, 2020, doi: 10.1155/2020/6622927.

[9] A. Yadav, C. K. Jha, and A. Sharan, "Optimizing LSTM for time series prediction in Indian
stock market," Procedia Computer Science, vol. 167, pp. 2091-2100, 2020. Doi:
10.1016/j.procs.2020.03.257.

[10] M. Nabipour, P. Nayyeri, H. Jabani, A. Mosavi, E. Salwana, and S. S., "Machine Learning
for Stock Market Prediction," Entropy, vol. 22, no. 8, p. 840, 2020. Doi: 10.3390/e22080840.
[11] H. N. Bhandari, B. Rimal, N. R. Pokhrel, R. Rimal, K. R. Dahal, and R. K. C. Khatri,
"Predicting stock market index using LSTM," Machine Learning with Applications, vol. 2022,
Article ID 100320, 2022, doi: 10.1016/j.mlwa.2022.100320.

[12] R. Qiao, W. Chen, and Y. Qiao, "Prediction of stock return by LSTM neural network,"
Applied Artificial Intelligence, vol. 36, no. 01, Article ID e2151159, 2022, doi:
10.1080/08839514.2022.2151159.

[13] Siami-Namini, S., & Namin, A. S. (2018). Forecasting economics and financial time series:
Arima vs. lstm. arXiv preprint arXiv:1803.06386.

[14] Roondiwala, M., Patel, H., & Varma, S. (2017). Predicting stock prices using LSTM.
International Journal of Science and Research (IJSR), 6(4), 1754–1756.

[15] Yu, P., & Yan, X. (2019). Stock price prediction based on deep neural networks. Neural
Computing and Applications, 32, 1609–1628

[16] Qiu M, Song Y. Predicting the direction of stock market index movement using an optimized
artifcial neural network model. PLoS ONE. 2016;11(5):e0155133.

[17] Shen J. Thesis, “Short-term stock market price trend prediction using a customized deep
learning system”, supervised by M. Omair Shafq, Carleton University. 2019.

[18] S. C. Nayak, B. B. Misra, and H. S. Behera, "Impact of Data Normalization on Stock Index
Forecasting," Artificial Intelligence in Finance, vol. 6, no. 2, pp. 98–114, 2020.

[19] Wang X, Lin W. Stock market prediction using neural networks: does trading volume help in
short-term prediction?. n.d

[20] Dixon, Matthew, Diego Klabjan, and Jin Hoon Bang. (2017) “Classification-based financial
markets prediction using deep neural networks.” Algorithmic Finance 6: 67–77.

[21] Sezer, Omer Berat, Murat Ozbayoglu, and Erdogan Dogdu. (2017) “A Deep Neural-Network
Based Stock Trading System Based on Evolutionary Optimized Technical Analysis Parameters.”
Procedia Computer Science 114: 473-480

[22] Olah, Christopher. (2015) Understanding LSTM Networks [Blog post]. Retrieved from
http://colah.github.io/posts/2015-08-UnderstandingLSTMs/

[23] Hiransha, M, Gopalakrishnan E. A., Vijay Krishna Menon, and Soman K. P. (2018) “NSE
Stock Market Prediction Using Deep-Learning Models.” Procedia Computer Science 132: 1351-
1362.

[24]Fischer, Thomas, and Christopher Krauss. (2018) “Deep learning with long short-term
memory networks for financial market predictions.” European Journal of Operational Research
270 (2): 654-669

[25] Zhang, G. Peter. (2003) “Time series forecasting using a hybrid ARIMA and neural network
mode.” Neurocomputing 50 : 159-175.

[26] Murkute, Amod, and Tanuja Sarode. (2015) “Forecasting market price of stock using artificial
neural network.” International Journal of Computer Applications 124 (12) : 11-15.

[27] L. Zhang, F. Wang, B. Xu, W. Chi, Q. Wang, and T. Sun, “Prediction of stock prices based
on LM-BP neural network and the estimation of overfitting point by RDCI,” Neural Computing
and Applications, vol. 30, no. 5, pp. 1425–1444, 2018.
[28] S. Hochreiter and J. Schmidhuber, “Long short-term memory,” MIT Press, vol. 9, no. 8, pp.
1735–1780, 1997.

[29] B. Svetlana and T. Ioannis, “An ensemble of LSTM neural networks for high-frequency stock
market classification,” Journal of Forecasting, vol. 38, no. 6, pp. 600–619, 2019.

[30] Y. Xue, C. Wang, and C. Miao, “Research on financial assets transaction prediction model
based on LSTM neural network,” Neural Computing and Applications, vol. 1, 2020.

[31] V.-D. Ta, C.-M. Liu, and D. A. Tadesse, “Portfolio optimization-based stock prediction using
long-short term memory network in quantitative trading,” Applied Sciences, vol. 10, no. 2, pp.
437–457, 2020.