ISSN (Online) 2581-9429

IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)
International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal
Impact Factor: 7.301 Volume 3, Issue 1, November 2023

Review Paper Enhancing Stock Price Prediction

using Optimized LSTM Networks
Gaurav Chaudhari, Mansi Ghive, Mrs. Nikita Khawase
Department of Artificial Intelligence & Data Science
ISBM College of Engineering, Nande, Pune, India
[email protected] and [email protected]
[email protected] and [email protected]

Abstract: Predicting stock market prices is challenging due to the stochastic and nonlinear nature of
financial time series data. A multi-step stock price forecasting model based on long short-term memory
(LSTM) is presented in this study. Recurrent neural networks of the LSTM type are excellent at identifying
long-term dependencies in time series. The models are trained and tested using historical stock price data
that is publicly available. Prior to being used as samples and labels for supervised learning, the data is
preprocessed to standardize its characteristics. An ideal architecture for an LSTM network is created by
methodical experimentation assessing performance in different configurations. Predictive capability is
evaluated using quantitative measures such directional correctness, mean absolute error, and root mean
squared error.
The LSTM model is compared with machine learning models such as support vector regression and
statistical techniques such as ARIMA. The outcomes show that, on a variety of assessment measures, the
deep learning LSTM technique performs better than the comparison models. The LSTM model can detect
latent pricing patterns efficiently and forecast future price changes with a high degree of accuracy. To
assess the effects of batching, model hyperparameters, and training window size, further experiments are
carried out. The research offers insights into the best model building techniques and demonstrates how
LSTM networks may be used to financial forecasting challenges. The suggested method can help traders
and investors by offering useful insights and indications based on anticipated stock prices.

Keywords: Stock, Forcasting, LSTM, Historical, Standardize, Predictive

I. INTRODUCTION
A crucial component of the world financial system, the stock market sees billions of dollars traded every day on stock
exchanges throughout the globe. Scholars, analysts, and investors have long sought to precisely forecast future market
patterns and stock prices. Recently, on financial time series forecasting challenges, machine learning algorithms have
shown encouraging outcomes. Specifically, deep learning techniques like as Long Short-Term Memory (LSTM)
networks have been useful for simulating sequences of stock prices.
Prior academic studies have extensively explored stock prediction using statistical models like ARIMA as well as
machine learning models like Support Vector Regression and Random Forests [1-3]. The availability of large amounts
of computational power and financial data has made deep neural networks useful for stock forecasting. LSTM networks
have been used in several research to forecast future stock values, and they have proven to perform better than other
methods. Nevertheless, rather than real-time analysis and trading, the majority of earlier work has concentrated on the
offline evaluation of prediction models.
Using Python frameworks such as Dash, this research creates an interactive web-based dashboard for real-time stock
price prediction and signal production. For more flexible examination, toggleable and customizable technical indicators
are used. The dashboard's prediction engine, an LSTM model, is trained using historical stock data. Trading signals are
created programmatically using indicator thresholds and aggregations to facilitate real-world application. Retail traders
may access an easy-to-use visualization interface on any device with the dashboard.

Copyright to IJARSCT DOI: 10.48175/568 127

www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)
International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal
Impact Factor: 7.301 Volume 3, Issue 1, November 2023

This work's primary contributions are:

1) An end-to-end program that prioritizes useful functionality over purely theoretical analysis. Using symbols that the
user provides, the dashboard provides trading indications and recommendations in real time.
2) Using up-to-date data, several configurable technical indicators are implemented. The original price data may be
displayed and overlaid with the indicators.
3) A multi-step forecasting optimized LSTM model was trained and assessed using a recent stock dataset. Comparing
quantitative indicators to statistical and machine learning standards, they perform better.
4) Trading signal creation using empirical reasoning and the sum of several indicator levels. This offers precise
instructions for entrances and exits.
5) An intuitive online interface for interactive analysis that is available on PC and mobile devices. Flexible usage is
made possible by price data, graphs, tooltips, and custom symbol search.

II. LITERATURE SURVEY

Stock market prediction is an active area of research in financial forecasting. Deep neural networks and other machine
learning techniques have gained popularity for stock price modeling due to the availability of large financial datasets.
Recurrent neural networks, such as Long Short-Term Memory (LSTM) networks, have shown remarkable efficacy in
processing sequential data, such as stock prices.
The first work that showed the potential of LSTM models for stock prediction was done by Moghar and Hamiche
(2020). They constructed a simple LSTM architecture using the daily opening prices of five major corporations.
Between 50 and 60 percent directional accuracy was attained across the stocks. Their findings indicated that LSTMs are
a promising research strategy.[1]
Subsequent work focused on improving the input features and model architecture. Zhang (2022) combined price data
with technical indicators such as moving averages. Additionally, they experimented with stacked LSTMs and batch size
optimization among other hyperparameters. The S&P 500 index's directional accuracy increased to 60% as a result. For
optimal performance, the study stressed the significance of feature engineering and parameter tuning[2].
Shaikh and Syed(2022)conducted a comprehensive analysis of different LSTM architectures. They assessed dropout
regularization, memory unit count, and the relative shallowness and depth of LSTMs. With National Stock Exchange
(NSE) stocks, they obtained a directional accuracy of 63% through methodical experiments. Their work offers helpful
guidelines for designing LSTM models[3].
By training a 3-layer stacked LSTM model, Talati et al. (2022) expanded on these findings. The accuracy on major
NSE stocks was improved to 63-65% by extracting hierarchical feature representations. The authors draw the
conclusion that complex data patterns overlooked by shallow models can be found by deep LSTMs[4].
Studies with a sector focus have also demonstrated promise. Pramod et al. (2021) used LSTMs with technical indicators
to forecast IT stock prices. With 60% accuracy, they were able to predict short-term trends, proving that tuned LSTMs
can be helpful even in specialized fields[5].
The significance of hyperparameter optimization is emphasized in more recent work by Venikar et al. (2022). They
methodically adjusted parameters like batch size, neurons, etc. in place of standard values. For an Indian oil company,
this increased the directional accuracy to about 62%. The findings demonstrate that the LSTM's default settings might
not be ideal [6].

Copyright to IJARSCT DOI: 10.48175/568 128

www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)
International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal
Impact Factor: 7.301 Volume 3, Issue 1, November 2023

III. PROPOSED SYSTEM

Fig:-System Architecture
The purpose of the suggested system is to use Long Short-Term Memory (LSTM) networks to generate signals and
build a strong framework for real-time stock price prediction. The goal of this system is to meet the critical demand for
precise forecasting in the fast-paced, fiercely competitive world of stock trading.
Long Short-Term Memory (LSTM) networks are a type of recurrent neural network (RNN) that we are using in this
proposed system to predict stock prices. Because LSTM networks excel at processing data sequences, they are
especially useful for time series prediction tasks like stock price forecasting. Long-term dependencies in the data can be
captured by them, which is essential for modeling intricate financial patterns.

Components;-
 Data Collection and Preprocessing
 LSTM Model Training
 Real-Time Dashboard
 Continuous Monitoring and Improvement

IV. CONCLUSION
This project demonstrated the effectiveness of Long Short-Term Memory (LSTM) networks for stock market
forecasting. To forecast the closing prices of stocks in the future, an LSTM model was created. Using historical price
data, the model was trained and evaluated.

Copyright to IJARSCT DOI: 10.48175/568 129

www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)
International Open-Access, Double-Blind, Peer-Reviewed, Refereed, Multidisciplinary Online Journal
Impact Factor: 7.301 Volume 3, Issue 1, November 2023

The findings demonstrate the LSTM model's ability to predict short-term stock price trends with accuracy. On the test
data, the model's overall directional accuracy was about 62%. This suggests that the situation is much better than
random chance.
Using regularization strategies, hyperparameter tuning, and an optimized model architecture are some of the main
elements that made the model successful. On the training set, methods such as dropout and recurrent dropout inhibited
overfitting. The generalization performance was enhanced by the methodical tuning of hyperparameters such as batch
size, epochs, and learning rate.
The study validates previous research indicating that long short-term memory (LSTM) systems are excellent choices for
financial time series forecasting. LSTMs are able to identify significant patterns in stock prices by capturing intricate
temporal relationships.
Still, there's room to improve the model's performance even more. By employing deeper architectures, adding more
engineered indicators, and fusing LSTMs with other cutting-edge models, the accuracy may be increased. Real-time
predictive analytics can be made possible by deploying the model via APIs.
All things considered, this work contributes to the mounting body of research showing that deep learning methods can
greatly expand the possibilities for stock market analysis. An effective and reliable method for utilizing financial data
for intelligent forecasting systems is through the use of LSTM networks.

REFERENCES
[1].Moghar, A., and M. Hamiche. "Stock market prediction using LSTM recurrent neural network." Procedia Computer
Science 170 (2020): 1168-1173.
[2].Zhang, Ruobing. "LSTM-based stock prediction modeling and analysis." Proceedings of the 2022 7th International
Conference on Financial Innovation and Economic Development (ICFIED 2022). Atlantis Press, 2022.
[3].Shaikh, Shoieb Abubaker, and Rouf Syed. "Stock Market Prediction Using LSTM Technique." International Journal
for Research in Applied Science & Engineering Technology 10.4 (2022): 3178-3184.
[4]. Talati, Drashti, et al. "Stock Market Prediction Using LSTM Technique." International Journal for Research in
Applied Science & Engineering Technology 10.6 (2022): 1820-1828.
[5]. P. M., Shastry, M. S., and B. S., Pramod. "Stock Price Prediction Using LSTM." TEST ENGINEERING AND
MANAGEMENT 83 (2021): 5246-5251.
[6]. Venikar, Isha, Jaai Joshi, and Harsh Jalnekar. "Stock Market Prediction Using LSTM." International Journal for
Research in Applied Science & Engineering Technology 10.12 (2022): 920-924 (ISSN: 2321-9653).

Copyright to IJARSCT DOI: 10.48175/568 130

www.ijarsct.co.in