Stock Price Prediction Using Machine

Learning

Project report submitted

in partial fulfillment of requirement for the degree

of

Bachelor of Technology
in
Computer Science & Engineering
Name of the students Name of
Supervisor:
Sumit Raj Sikarwar (1905110100115) Er. Jayati Krishna
Goswami
Anurag Sharma (1905110100027) Assistant Professor

GL BAJAJ GROUP OF INSTITUTIONS,

MATHURA
Session: 2022-23
 Department of Computer Science & Engineering
CERTIFICATE

This is to certify that:

Mr. Sumit Raj Sikarwar (1905110100115), Mr. Anurag Sharma
(1905110100027) has worked on Stock Price Prediction Using
Machine Learning. This project is part of a partial fulfillment of
requirement for the degree of Bachelor in Technology in Computer
Science Engineering.

To the best of our knowledge and belief, this is the original work and
has not been submitted for any other degree elsewhere.

Date :
Place:

Signatures:

(Er. Jayati Krishna Goswami)

CSE

(Mr. Ramakant Baghel)

HOD CSE
 DECLARATION

I hereby certify that the work which is being presented in B. Tech.

Project Report entitled “Stock Price Prediction Using Machine
Learning”, as partial fulfillment of the requirement for the degree of
Bachelor of Technology in Computer Science Engineering, submitted to
the Department of Computer Science Engineering of GL BAJAJ Group
of Institutions, NH#2- Mathura-Delhi Road, PO-Akbarpur, Mathura-
281001 (UP), is an authentic record of my own work carried out during a
period from October 2022 to April 2023 under the supervision of Mr.
Vipul Kadam and Mr. Jayati Krishna Goswami in the Computer Science
Department.

The matter presented in this project report in full or part, has not been
submitted by me for the award of any other degree elsewhere and is free
from plagiarism.

Name of Candidates:

Sumit Raj Sikarwar

(1905110100115)

Anurag Sharma
(1905110100027)
 ACKNOWLEDGEMENTS

We are thankful to and would like to pay our special regards to G.L
BAJAJ GROUP OF INSTITUTIONS for providing required resources for
this work. I wish to express my sincere appreciation to my supervisors,
their assistance was a milestone in completion this project. Prof Mr.
Jayati Krishna Goswami persistently guided me in planning of work and
Prof. Mr. Ramakant Baghel intelligently solved my queries. Without their
support, a significant portion of project would not be possible in
constrained time.

We are also thankful to and so must also appreciate our families and
friends for helping us survive all the stress throughout the year. To our
parents, for supporting us both on and off the water. We would like to
thank you all mentioned and other people who have helped us directly or
indirectly for pushing us farther than we thought we could.

Signature & Name of Student(s):

Sumit Raj Sikarwar Anurag Sharma

 Abstract

One of the main objectives of this project is to improve the quality of

output of stock market predictions by using the value of stocks as an

input. Researchers have been investigating different methods to

effectively predict the stock market price, which can provide useful

insights for traders and investors about the future trends and conditions of

the market. Some of the traditional methods that have been used for this

purpose are artificial neural network, which is a technique that can find

hidden patterns and classify the data that is used for forecasting the stock

market.

However, this project proposes a different method that does not depend

on fitting the data to a specific model, but rather identifies the latent

dynamics that exist in the data using machine learning architectures. In

this work, we use two types of machine learning architectures: Long

Short-Term Memory (LSTM) and Convolutional Neural Network (CNN)

to predict the prices of companies listed on NSE and differentiate their

performance. We apply a sliding window approach for long term

predictions and measure the performance using root mean square error as

a metric.
 LIST OF FIGURES

FIG. NO
TITLE PAGE NO
3.1 Proposed Work Flow 11
3.2
LSTM Architecture 12
3.3
CNN Architecture 14
4.1
Stock Dataset Information 20
4.2
Read Dataset 20
4.3
Graph of Closing-Price history 21
4.4
Data-scaling 21
4.5.
Long Short Term Memory Summary 22
4.6
CNN Summary 22
4.7
Training Process 23
4.8
Predicted Close Price 23
4.9
Plot for Real vs Predicted value for GOOGLE using 24
Long Short Term Memory (LSTM)
4.10
Plot for Real vs Predicted value for GOOGLE using 25
CNN
4.11 Plot for Real vs Predicted value for INFY using 25
LSTM

4.12 Plot for Real vs Predicted value for INFY using 25

CNN
4.13 Plot for Real vs Predicted value for T-MOBILE US using
LSTM
4.14 Plot for Real vs Predicted value for T-MOBILE US using 26
CNN
4.15 Plot for Real vs Predicted value for TESLA using LSTM 26

4.16 Plot for Real vs Predicted value for TESLA using CNN 27
4.17
Plot for Real vs Predicted value for TWITTER using LSTM 27
4.18
Plot for Real vs Predicted value for TWITTER using CNN 27
 LIST OF ABBREVIATIONS

1. ARIMA - Auto Regressive Integrated Moving Average

2. ANN - Artificial Neural Network
3. BSE - Bombay Stock Exchange
4. CNN - Convolutional Neural Network
5. CPU - Central Processing Unit
6. CRNN - Convolutional Recurrent Neural Network
7. CSV - Comma Separated Values
8. EMA - Exponential Moving Average
9. FEX - Foreign Exchange
10. GPU - Graphics Processing Unit
11. GSFC - Gujarat State Fertilizers & Chemicals Limited
12. KSE - Karachi Stock Exchange
13. LSTM - Long Short-Term Memory
14. MA - Moving Average
15. MAE - Mean Absolute Error
16. MLP - Multi-Layer Perceptron
17. MLR - Multiple Linear Regression
18. MSE - Mean Square Error
19. NASDAQ - National Association of Securities Dealers
Automated Quotations
20. NSE - National Stock Exchange
21. P/E Ratio - Profit per Earning ratio
22. RBF - Radial Basis Function
23. RMSE - Root Mean Square Error
24. RNN - Recurrent Neural Network
25. SLP - Single Layer Perceptron
26. SVM - Support Vector Machine
 Chapter 1. Introduction

1.1Stock Price Prediction

Given the soaring profitability of the stock market, it has emerged as an

immensely popular investment avenue. As trading and investing activities
continue to escalate, individuals are actively exploring techniques and
tools to maximize their gains while minimizing associated risks. In the
Indian stock market, the National Stock Exchange (NSE) and the Bombay
Stock Exchange (BSE) stand out as the primary platforms for trading.
Notably, the Sensex and Nifty serve as significant market indexes in
India. Due to the dynamic nature of stock prices, predicting market trends
poses a considerable challenge.

Over the years, various forecasting models have been devised and
successfully implemented in the realm of market price prediction.
Generally, this classification could be done as :
1. Time-series analysis
2. Fundamental analysis
3. Technical analysis

Time-Series Analysis

Forecasting entails the assessment of forthcoming outcomes or results by

scrutinizing historical data. This practice finds applications across diverse
domains such as industry & business, economics and finance, and
environmental science. Forecasting problems can be categorized into two
main types:

1. Long-term forecasting: This involves estimating outcomes that extend

beyond a period of two years.
 2. Medium-term forecasting: Here, the focus lies on estimating outcomes
for a duration ranging from one to two years.

In addition to long-term and medium-term forecasting, short-term forecasting

focuses on estimating outcomes over shorter durations, ranging from weeks and
months to days, minutes, or even a few seconds. The analysis of time series
encompasses various forecasting problems [1].

A time series is essentially a sequential arrangement of observations for a specific

variable, such as stock prices in our case. Time series data can be classified as either
multivariate or univariate. In univariate data, only a particular stock is considered,
whereas multivariate data involves multiple companies for different time instances.

Time series analysis offers several advantages for examining trends, patterns, cycles,
or periods within the available data. This analysis is particularly valuable for
identifying early indications of bullish or bearish trends in the stock market, aiding in
making wise investment decisions.

Furthermore, pattern analysis within a specific period plays a crucial role in

categorizing the best-performing companies. These factors highlight the significance
of forecasting and time series analysis as vital research areas.

Fundamental Analysis

Fundamental analysts primarily focus on assessing the underlying factors

that drive a stock's value. They thoroughly examine a company's historical
performance and the reliability of its financial records. To gauge the
validity of a stock, fundamental analysts utilize various performance
indicators, such as the P/E ratio. Among the well-known fundamental
analysts, Warren Buffett stands out as a prominent figure.

The goal of fundamental analysis in the stock market is to determine the

intrinsic value of a stock, which can then be compared to its market price.
By doing so, analysts can ascertain whether a stock is undervalued or
overvalued. Estimating the true value of a stock involves employing
various methods, but they all revolve around the same principle: a
company's worth is based on its future profits. These future profits are
 discounted to their present value. This principle assumes that a business's
primary focus should be on generating profits. In contrast to technical
analysis, fundamental analysis is considered a longer-term approach.

Fundamental analysis is built on the belief that society requires capital for
progress, and if a company performs well, it should be rewarded with
additional capital, leading to an increase in stock price. This type of
analysis is commonly employed by fund managers because it relies on
sensible, objective evaluation using publicly available data, such as
financial statements.

Another aspect of fundamental analysis extends beyond bottom-up

business analysis. It incorporates a top-down approach that begins with
analyzing the global economy, followed by country analysis, sector
analysis, and finally, company-level analysis.

Technical Analysis

The practitioners known as chartists, esteemed as technical analysts, do

not confine their contemplations to the fundamental facets of a company.
Instead, their predilections lie predominantly in the realm of perusing and
dissecting the tapestry of past price trends, which can be discerned as a
manifestation of time series analysis, in order to prophesy the future
undulations of a stock's price.

These adept individuals employ an array of intricate patterns, such as the

revered head and shoulders formation or the graceful cup and saucer
design, as their brushstrokes on the canvas of technical analysis.
Furthermore, they deftly wield an assortment of techniques and indicators,
including oscillators, the resolute exponential moving averages (EMA),
steadfast support and resistance levels, resounding momentum indicators,
and the sonorous volume indicators. Amongst their repertoire, the elegant
artistry of candlestick patterns, once harnessed by astute Japanese rice
merchants, has now permeated the very fabric of technical analysis.

Technical analysis, with its gaze keenly trained on the ephemeral realms
of short-term trading pursuits, finds its veritable haven within the dynamic
domains of commodities and forex markets. It is within these arenas that
traders fervently fixate upon the capricious fluctuations of price, seeking
to seize fleeting opportunities. And in this endeavor, they adhere to a set
of cardinal principles that govern the sagacious minds of technical
analysts. Firstly, they ardently uphold the belief that all salient
information pertaining to a company is already encapsulated within the
ethereal realm of its stock price. Secondly, they keenly observe the
rhythmic undulations of price movements, recognizing the cadence of
trends that courses through the market's veins.

And lastly, technical analysts place their faith in the unyielding tenet that
the annals of history are a prism through which price patterns reverberate,
echoing the symphony of market psychology and human behavior.

It is of paramount importance to acknowledge the clear distinction that

sets technical analysis apart from its counterpart, fundamental analysis.
While the latter delves deep into the multifaceted layers of a company's
fundamental fabric, technical analysis shines its resplendent light solely
upon the resolute tapestry woven by price and volume data, relinquishing
the fundamental aspects to the purview of its scholarly sibling.
 1.2Applications
1. Businesses
2. Companies
3. Insurance companies
4. Government Agencies
5. This application is much helpful for stock investors, sellers, buyers, and
brokers too.

1.3Objectives
Stock market prediction involves the endeavor to forecast the future value
of a company's stock or other financial investments traded on the stock
exchange. Successfully estimating the upcoming price of a stock can yield
significant profits, enabling investors to make wise investment decisions
and generate substantial returns. By employing various analysis
techniques and methodologies, investors strive to anticipate market trends
and make informed choices that lead to profitable outcomes. Effective
stock market prediction plays a crucial role in guiding investment
strategies and maximizing potential profits.

1.4 Motivation

The primary motivation behind stock price prediction is to ascertain the

future price of a stock. This motivation extends across multiple domains,
including business, industry, environmental science, finance, and
economics. Understanding the future value of a company's stock holds
significant importance in making informed investment decisions. Various
techniques, analyses, and models are employed to estimate and determine
the potential future value of a stock. By leveraging these predictions,
investors can gain insights into the potential profitability of investing in a
particular company's stock.
1.4 Report Organization
Chapter 2 of the research study encompasses a literature survey that offers
a comprehensive summary of individual papers related to the topic at
hand. It provides an overview of existing research and scholarly
contributions from various authors, highlighting their key findings and
methodologies.

In Chapter 3, an in-depth overview is presented regarding the existing

work conducted for stock price prediction utilizing Long Short-Term
Memory (LSTM) and Convolutional Neural Network (CNN) techniques.
The chapter explores the studies and research that have employed these
specific methodologies, discussing their approaches, models, and
outcomes.

Chapter 4 focuses on the implementation of the stock price prediction

model and presents the results obtained. It delves into the tools and
technologies employed in the implementation process, providing details
about the dataset utilized for training and testing the model.

The concluding Chapter 5 offers a comprehensive summary of the

research findings on stock price prediction. It discusses the key insights
and conclusions drawn from the study, highlighting the implications and
significance of the results obtained. Additionally, this chapter outlines
future directions and potential areas of further exploration in the field of
stock price prediction, providing a glimpse into the author's intended
future work.
 Chapter Second- Literature Survey

Stock Price Forecasting Using Data From Yahoo Finance and

Analyzing Seasonal and Non-seasonal trend:
Publication Year : 2018

Author: Jai Jagatwani, Hardik Sachdeva, Manav Gupta, Alka Singhal

Journal Name: 2018 IEEE
Summary: The proposed work aims to examine the relationship between
two existing time series algorithms, namely ARIMA and Holt-Winters,
and their applicability in predicting stock prices. The primary objective is
to analyze a risk-free range of stock prices for investment purposes, with
the goal of achieving improved accuracy in the prediction model. By
combining these two different time series analysis models, the aim is to
generate a range of price predictions for stock market investors.

One notable advantage of these models is their simplicity and reliance on

past stock prices for forecasting values, whether the data is non-seasonal
or seasonal in nature. The models do not involve complex computations
and primarily use historical stock price data for estimation.

However, it is important to acknowledge some limitations of the proposed

work. The study does not take into consideration external factors such as
news related to new market strategies or media releases that could
potentially impact stock prices. These external factors, which can
significantly influence stock prices, are not incorporated into the analysis.
It is crucial to recognize the impact of such events and their potential to
affect the accuracy and reliability of the stock price predictions generated
by the proposed models.
 Stock Market Prediction Using Machine Learning :
Publication Year: 2018

Author: Ishika Parmar, Ridham Arora, Lokesh Chouhan, Navanshu Agarwal,

Shikhin Gupta, Shirish Saxena, Himanshu Dhiman
Journal Name: 2018 IEEE
Summary: The focus of this study revolves around the utilization of
regression and LSTM-based machine learning techniques for stock price
forecasting. The factors considered in the analysis include open price,
close price, low price, high price, and trading volume. The main objective
of the study is to enhance the accuracy and reliability of predicting future
stock prices using these machine learning methods.

The results of the study demonstrate that the LSTM algorithm yielded
positive outcomes, showcasing higher accuracy in predicting stock prices.
By leveraging the capabilities of LSTM-based machine learning, the
study aimed to improve the precision of forecasting the stock prices of a
company. The findings suggest that employing LSTM in the prediction
process led to more reliable and accurate stock price forecasts compared
to other methods used in the study, such as regression.

Overall, the study emphasizes the potential of machine learning

techniques, particularly LSTM, in enhancing the accuracy and reliability
of stock price predictions. By considering multiple factors and utilizing
advanced algorithms, the study contributes to the field of stock market
analysis and forecasting.
 Multi-Category Events Driven
Stock Price Trends

Publication Year: 2018

Author: Youxun Lei, Kaiyuve Zhou, Yuchen Liu

Journal Name: 2018 IEEE
Summary: The focus of this paper is on utilizing multi-category news
events as features to develop a stock price trend prediction model. These
multi-category events are derived from a predefined word dictionary for
feature extraction. The study employs both neural networks and Support
Vector Machine (SVM) models to analyze the relationship between stock
price movements and specific multi-category news.

The experimental results indicate that the use of predefined multi-

category news events as features leads to improved performance
compared to the baseline bag-of-words feature approach in predicting
stock price trends. The study demonstrates that incorporating multi-
category news events as features provides more valuable information for
predicting stock price movements.

Furthermore, the results indicate that the model performs better in short-
term prediction rather than long-term prediction based on the findings of
this study. This implies that the model's accuracy is higher when
predicting stock price trends over a shorter time horizon.

Overall, this research highlights the significance of utilizing multi-

category news events as features in stock price trend prediction models.
The improved performance compared to traditional bag-of-words
approaches and the focus on short-term prediction contribute to advancing
the understanding and application of stock market forecasting.
 Share Price Prediction using Machine Learning Technique:
Publication Year: 2018

Author: Jivan B, Naresh E., Vijay Kumar B.P, Prashant Kambly

Journal Name: 2018 IEEE
Summary: The main focus of this paper is on predicting stock prices
using Long Short-Term Memory and Recurrent Neural Networks (RNN).
The study utilizes these machine learning algorithms to forecast the stock
value based on data from the National Stock Exchange (NSE),
considering various factors such as current market price, price-earning
ratio, base value, and other unidentified events.

The efficiency of the model is evaluated by comparing the predicted data

with the actual data using an RNN graph. The results demonstrate that the
model is able to predict stock prices that closely align with the actual
prices. The model captures detailed features and employs different
strategies to make accurate predictions.

To train the model, NSE data from the internet is used, and the input is
organized and grouped based on user configuration. The RNN-based
architecture proves to be highly effective in forecasting stock prices, as it
adjusts the configuration accordingly. Additionally, the model
incorporates the backpropagation mechanism to handle data collection
and grouping, ensuring that the data remains separate and avoids any
mixing.

Overall, the study highlights the effectiveness of using LSTM and RNN
models for stock price prediction, leveraging various factors and
strategies to improve accuracy. The model's ability to capture detailed
features and adapt to different configurations contributes to its success in
forecasting stock prices.
 Stock Market Prediction Using
Machine Learning Techniques:
Publication Year: 2016

Author: Mehak Usmani, Syed Hasan Adil, Kamran Raza, Syed Sad Azhar Ali
Journal Name: 2016 IEEE
Summary: The primary objective of this study is to forecast the market
performance of the Karachi Stock Exchange (KSE) based on day closing
using machine learning algorithms. The model predicts whether the
market performance will be positive or negative, utilizing a range of input
attributes. These attributes include oil rates, gold and silver rates, interest
rates, foreign exchange (FEX) rates, news, and social media feed.

Several machine learning algorithms, namely Single Layer Perceptron

(SLP), Multi-Layer Perceptron (MLP), Radial Basis Function (RBF), and
Support Vector Machine (SVM), are compared in this study. The MLP
algorithm, specifically the multi-layer perceptron, demonstrated the best
performance compared to other methods.

The study identifies that the most influential feature in predicting the
market performance was the oil rate attribute. The findings validate that
machine learning techniques have the capability to predict the stock
market's performance. The Multi-Layer Perceptron algorithm achieved a
70% accuracy in correctly predicting market performance.

Overall, this research highlights the potential of machine learning

algorithms, particularly the MLP model, in forecasting stock market
performance. By analyzing a range of attributes and utilizing advanced
algorithms, the study provides valuable insights into predicting market
trends and enhancing investment decision-making processes.
 Forecasting stock price in two ways based
on LSTM neural network:

Publication Year: 2019

Author: Jingyi Du, Qingli Liu, Kang Chen, Jiacheng Wang

Journal Name: 2019 IEEE
Summary: In this study, the LSTM neural network is employed to predict
Apple stocks. The model is evaluated using both single feature input
variables and multi-feature input variables to examine its forecasting
performance on stock time series data.

The experimental results demonstrate that the model achieves a high level
of accuracy when using multi-feature input variables, with an accuracy
rate of 0.033. This indicates that the model's predictions closely align with
the actual stock values and meet the desired forecasting objectives.

On the other hand, when using univariate feature input, the model exhibits
a higher predicted squared absolute error of 0.155. This suggests that the
model's performance is inferior compared to the multi-feature variable
input scenario. The use of multiple input features allows the model to
capture a broader range of information and patterns, resulting in improved
accuracy in stock price prediction.

Overall, the study highlights the effectiveness of the LSTM neural

network in predicting Apple stock prices. The results emphasize the
importance of utilizing multiple input features to enhance the model's
forecasting accuracy in stock time series analysis.
Share Price Trend Prediction Using CRNN with
LSTM Structure:

Publication Year: 2018

Author: Shao-En Gao , Bo-Sheng Lin ,Chuin-Mu Wang

Journal Name: 2018 IEEE
Summary: The stock market plays a significant role in the entire financial
market, and predicting stock price volatility is a crucial research topic.
This paper focuses on utilizing deep learning methods to estimate the
future stock prices based on historical stock information and approximate
the trend value of stock prices.

The paper incorporates a designed neural network that stores historical

stock data with a time scale and uses it to predict future stock prices. It
recognizes that the trend of stocks is often connected to the previous
information of stock prices. To address this, the paper proposes the use of
a convolutional recurrent neural network (CRNN) with memory
performance and integrates the Long Short-Term Memory (LSTM) to
improve the long-term dependency of traditional Recurrent Neural
Networks (RNNs).

The goal is to enhance the accuracy and stability of stock price prediction
by employing the RNN LSTM architecture. The paper conducts
experiments using ten sets of historical stock data and achieves an average
error rate of 3.449 Root Mean Square Error (RMSE).

Overall, this study demonstrates the effectiveness of utilizing deep

learning methods, specifically CRNN and LSTM, in predicting stock
prices and estimating their volatility. By incorporating historical stock
data and enhancing the long-term dependency of the neural network, the
paper contributes to improving the accuracy and stability of stock price
prediction.
Applying Long Short Term Memory Neural Networks for Predicting
Stock Closing Price:
Publication Year: 2017

Author: Tingwei Gao, Yueting Chai, Yi Liu

Journal Name: 2017 IEEE
Summary: The main objective of this paper is to evaluate the
effectiveness of combining Recurrent Neural Networks (RNNs) with
informative input variables to improve the accuracy of stock market
forecasting for the next-day market. The study focuses on using Long
Short-Term Memory (LSTM) and stock basic trading data to analyze and
predict stock prices.

The case study relies on data from Standard & Poor's (S&P500) and
NASDAQ. The researchers develop a stock prediction model that
leverages LSTM and compares it with other models for performance
evaluation. The main discovery of the case study is that their forecasting
system provides a more accurate prediction of the stock closing price for
the next day compared to the comparison models.

To demonstrate the utility of the system, the researchers test five different
models: Moving Average (MA), Exponential Moving Average (EMA),
Support Vector Machine (SVM), and LSTM. The target of prediction is
the closing value of the stock for the next day.

Overall, the study highlights the effectiveness of incorporating LSTM and

informative input variables in stock market forecasting. The results show
that the proposed system outperforms other models and provides a more
precise prediction of the next-day stock closing price. This research
contributes to advancing the field of stock market prediction and
emphasizes the utility of LSTM in financial forecasting.
Developing a Prediction Model for Stock Analysis:
Publication Year: 2017

Author: R.Yamini Nivethha, Dr. C.Dhaya

Journal Name: 2017 IEEE
Summary: The main objective of this study is to conduct a comparative
analysis of three algorithms: Multiple Linear Regression (MLR), Support
Vector Machine (SVM), and Artificial Neural Network (ANN). The study
aims to determine the effectiveness of these algorithms in predicting the
market price for the upcoming day, considering both monthly and daily
predictions. Additionally, sentiment analysis is incorporated to enhance
the prediction accuracy using the best-performing algorithm.

Among the three algorithms, Multiple Linear Regression is found to be

less developed. It calculates the correlation between volume and the stock
price. On the other hand, deep learning algorithms, such as SVM and
ANN, demonstrate more advanced performance in stock price prediction.

The results of the study suggest that deep learning algorithms outperform
MLR in terms of prediction accuracy. The SVM algorithm also shows
promising results. By leveraging sentiment analysis and selecting the most
effective prediction algorithm, the study enhances the accuracy of stock
price forecasting.

In summary, this research highlights the comparative analysis of MLR,

SVM, and ANN algorithms for stock price prediction. The findings
underscore the superiority of deep learning algorithms over MLR and
demonstrate the potential of sentiment analysis in improving prediction
accuracy.
Stock Price Prediction Based on Information Entropy and Artificial
Neural Network:
Publication Year: 2019

Author: Zang Yeze, Wang Yiying

Journal Name: 2019 IEEE
Summary: The stock market plays a crucial role in the financial system as it
enables investors to allocate funds to various firms. In this paper, a machine
learning framework is developed by combining information theory and
Artificial Neural Network (ANN) to support the activities and evolution of the
stock market.

The framework incorporates information entropy, which measures the

uncertainty and relevance of stock data, to capture non-linear causality and
facilitate time series modelling using ANN. The feasibility and effectiveness of
this machine learning framework are evaluated using stock prices from major
companies such as Amazon, Apple, Google, and Facebook.

The paper outlines a time series analysis method that combines information
theory and Long Short-Term Memory (LSTM) to model the dynamics of stock
prices. The integration of transfer entropy, which quantifies the information
flow between relevant variables, enhances the prediction accuracy of the LSTM
model. The results show a high correlation between the modelled and real stock
prices, although there may be slight differences in terms of Mean Absolute
Error (MAE) and Root Mean Square Error (RMSE).

Overall, this research demonstrates the successful application of a machine

learning framework that combines information theory and ANN for stock price
prediction. By integrating information entropy and transfer entropy into the
modelling infrastructure, the framework provides accurate predictions and
valuable insights into stock market dynamics.
 Summary of Literature Survey :

In our review of various approaches for stock price prediction, we have

highlighted the advantages and challenges associated with CNN & LSTM,
which are popular algorithms in this domain. While these algorithms have
shown promise in predicting stock prices, they do come with certain
limitations.

Some of the challenges mentioned include the requirement for a large

amount of training data, high computational costs, slower training without
GPU acceleration, and the dependence on previous information for
accurate predictions.

To overcome these challenges, a hybrid approach can be considered. A

hybrid approach combines multiple techniques or algorithms to leverage
their strengths and mitigate their weaknesses. For example, combining
CNN and LSTM with other machine learning algorithms or traditional
statistical methods can provide a more robust and accurate prediction
model.

Additionally, machine learning techniques, when applied using standard

tools and frameworks, have shown great potential in stock price
prediction. These methods have been found to outperform traditional
prediction methods and can provide highly accurate results.

It's important to note that while machine learning can offer powerful
prediction capabilities, it is still essential to carefully consider the quality
and relevance of the input data, as well as the appropriate preprocessing
and feature selection techniques.

In conclusion, a hybrid approach that combines different techniques and

leverages the strengths of machine learning algorithms can help address
the challenges associated with stock price prediction, leading to more
accurate and reliable predictions.
 Chapter Three - Design of Project Model
3.1 Overview of Existing Work

The proposed system for stock price prediction in this paper utilizes
machine learning techniques, specifically the LSTM algorithm, to
estimate the future value of stocks. LSTM is chosen as it has shown to
perform well in capturing the behavioral changes in stock prices over a
given period.

The dataset used for analysis is obtained from Yahoo Finance and consists
of approximately 900,000 records of stock prices and relevant data. The
data includes information such as date, symbol, open price, close price,
low price, high price, and volume. Only data for a single company is
considered in this study. The data is preprocessed by transforming it into a
data frame using the Pandas library in Python.

Normalization of the data is performed using the sklearn library in

Python, and the dataset is split into training and testing sets. The
experiment set, used for evaluation, is kept as 20% of the available
dataset.

The paper focuses on two architectures: the Regression-based Model and

LSTM. The Regression-based Model is employed to predict continuous
values based on given independent variables using a linear function.
LSTM, on the other hand, is capable of identifying changes in trends and
shows promising results in this proposed system.

The study concludes that LSTM is the best model for the proposed
methodology, indicating its ability to identify patterns and relationships
within the data. It acknowledges that the stock market may not always
follow a regular pattern or cycle, and the duration and existence of trends
can vary based on companies and sectors. By analyzing trends and cycles
using techniques like LSTM, investors can potentially obtain higher
profits by leveraging current information and analyzing various factors.

In summary, the proposed system utilizes machine learning techniques,

specifically LSTM, for stock price prediction, demonstrates the
importance of analyzing trends and cycles in the stock market and states
the benefits of using LSTM to capture relevant information and make
predictions.
 3.2 Proposed Work

Fig 3.1 Proposed-Workflow

The presented system comprises of five modules :

i. Dataset (Input for process)

ii. Preprocessing
iii. Data splitting
iv. Build & Model train LSTM & CNN.
v. Output as Predicted Result

Various attributes related to stock prices, such as open price, high price,
low price, close price, and adjusted close price, are extracted from a large
dataset. These attributes serve as input features for training the models.
To preprocess the data, techniques like normalization and one-hot
encoding are applied. Normalization is used to scale the numerical
features within a certain range, while one-hot encoding is employed to
convert categorical variables into binary vectors. The preprocessed data is
then divided into two sets: a training set and a testing set. The ratio of the
split is 80:20, meaning 80% of the data is used for training the models,
and the remaining 20% is reserved for evaluating their performance.
Three different approaches are used to train the models: LSTM, CNN, and
a combination of LSTM and CNN (LSTMCNN). These models are
trained using the training set, and their parameters are adjusted to
minimize the prediction errors.

To evaluate the performance of the models, the root mean square error
(RMSE) is calculated. RMSE measures the difference between the
predicted values and the actual values, giving an indication of how well
the models are able to estimate the stock prices.

By comparing the RMSE values obtained from the different approaches, it

is possible to assess the effectiveness of each model in predicting stock
prices. Lower RMSE values indicate better accuracy and performance.

In summary, the study involves preprocessing the dataset by applying

techniques like normalization and one-hot encoding. The data is then split
into training and testing sets. Three approaches, LSTM, CNN, and
LSTMCNN, are employed to train models using the training data. The
models are evaluated based on their RMSE values, providing insights into
their prediction accuracy.
 3.2-1 Working of LSTM Model

LSTM-Architecture

LSTM is a type of recurrent neural network (RNN) that addresses the

issue of long-term dependencies in sequential data. It retains information
over longer periods, making it effective for processing, predicting, and
classifying time-series data. Its specialized memory cells and gating
mechanisms allow it to selectively remember or forget information,
improving the accuracy of predictions. LSTM is widely used in various
fields, including natural language processing and stock price prediction.

Structure Of LSTM :
LSTM (Long Short-Term Memory) consists of a chain-like structure that
includes four neural networks and specialized memory cells called cells.
One of the key features of LSTM is the presence of memory cells, which
make decisions regarding storing, reading, writing, and forgetting
information.
A memory cell in LSTM is equipped with three main gates:
oInput gate: Controls the flow of new information into the memory cell.
oForget gate: Regulates which information remains stored in the memory
cell.
oOutput gate: Determines how the information in the memory cell is used
to compute the output.
These gates enable LSTM to effectively process and retain important
information over longer sequences, allowing it to capture and utilize long-
term dependencies in the data.
LSTM Has Several Major Applications :

 Handwriting-generation

 Image Captioning

 Language Modelling

 Machine Translation

 Question-Answering Chatbot
Working of CNN model :

CNN Architecture

 Layers of a CNN model:

A Convolutional Neural Network (CNN) model consists of several key layers that
collectively enable effective image processing and feature extraction. These layers
include:

1. Convolution: The convolution layer performs feature extraction by applying filters

to the input image. It convolves the filters across the image to capture important
spatial information and extract relevant features. The output is obtained in the form
of a feature map.

2. MAX Pooling: MAX pooling is a downsampling operation that reduces the spatial
dimensions of the feature map. It selects the maximum value within a local region,
thereby preserving the most salient features while reducing computational
complexity and enhancing translation invariance.

3. Dropout: Dropout is a regularization technique used during training. It randomly

drops out a fraction of neurons, making them inactive during forward and backward
propagation. This prevents over-reliance on specific neurons and encourages the
network to learn more robust and generalizable representations.

4. Flatten: The flatten layer reshapes the multidimensional feature map into a one-
dimensional vector. It converts the spatial structure into a sequential format,
preparing the data to be fed into a fully connected layer.

5. Dense: The dense layer, also known as a fully connected layer, is responsible for
learning complex relationships between the extracted features. Each neuron in this
layer is connected to every neuron in the previous layer, and the connections are
weighted. The dense layer performs linear operations followed by non-linear
activation functions.

6. Activation: The activation layer introduces non-linearity to the network. It applies

a specific activation function (such as the sigmoid function) to the outputs of the
previous layer. Activation functions transform the output values into a desired range
(e.g., between 0 and 1) and enable the network to model complex relationships and
make predictions.

In summary, the layers in a CNN model work in tandem to extract features, reduce
spatial dimensions, regularize the network, capture complex relationships, and
introduce non-linearity, ultimately enabling the model to learn and make predictions
based on input images.
 Chapter Four - Dataset, Implementation and
Result

4.1 Dataset Detail

The dataset utilized in this study encompasses a wide range of historical stock data
obtained from the National Stock Exchange (NSE). It provides a comprehensive and
detailed view of the daily performance and behavior of various stocks across
different sectors. The dataset covers multiple sectors, including Banking, Pharma,
Petroleum, Software, and Textiles, allowing for a diverse analysis of the stock
market.

For each stock, the dataset captures essential attributes that are crucial in
understanding its performance. These attributes include the opening price, which
represents the initial trading price of the stock at the beginning of the trading day.
The highest price denotes the maximum price at which the stock was traded during
the day, while the lowest price signifies the minimum price observed for the stock
during the same period.

The closing price indicates the final trading price of the stock at the end of the
trading day, providing valuable information about the overall market sentiment and
investor activity. The adjusted closing price takes into account any corporate actions
or events that might affect the stock's value, such as stock splits or dividends.

In addition to price data, the dataset also includes the trading volume of stocks,
which reflects the total number of shares traded for a particular stock during a given
period. This volume information is crucial in understanding market liquidity and the
level of investor participation in buying or selling a particular stock.

By incorporating such a comprehensive dataset, researchers and analysts can gain

insights into the historical trends, patterns, and behavior of stocks across different
sectors. This rich dataset forms the foundation for conducting in-depth analysis,
developing predictive models, and making informed investment decisions in the
stock market.
 4.2Technologies And Tools

4.2.1 PYTHON

The choice of Python as the programming language for this project was made
for several compelling reasons. Firstly, Python boasts a large and active
community of developers, making it easy to find solutions and resolve any
challenges encountered during the project. The availability of resources like
Stack Overflow ensures that assistance is readily available for troubleshooting
and problem-solving.

Python is widely regarded as one of the most popular languages for

scientific computing. It offers a plethora of powerful libraries and
packages specifically tailored for tasks in this domain. Packages such as
NumPy, Pandas, and SciPy provide efficient and well-documented tools
for data manipulation, analysis, and numerical computations. Leveraging
these packages significantly reduces the amount of code required to
accomplish complex tasks, leading to faster development and iteration.
Another advantage of Python is its readability and ease of use. The
language syntax allows for writing code that closely resembles pseudo
code, making it intuitive and straightforward to express algorithms and
ideas. This is particularly advantageous when working with tutorial
papers.

However, Python does have its drawbacks. Being a dynamically typed

language, it can sometimes lead to unexpected behavior, especially when
working with third-party packages that may have loosely defined return
types. This can introduce challenges and require additional testing and
experimentation to ensure the desired outcomes. The documentation for
certain packages may also lack clarity regarding method return types,
adding to the learning curve when using new libraries. Despite these
challenges, the benefits of using Python for this project outweighed the
drawbacks. The extensive availability of scientific computing packages,
the supportive community, and the readability of the language all
contributed to a smoother and more efficient development process.
 4.2.2 NUMPY
NumPy is a powerful Python package that provides essential tools and
functionality for scientific computing and numerical operations. It is
considered the core library for scientific computing in Python. Some key
features and capabilities of NumPy include:

1. N-dimensional array object: NumPy's ndarray is a powerful data

structure that allows efficient storage and manipulation of arrays. It
provides a multidimensional container for homogeneous data, such as
numerical values, and enables operations on these arrays.

2. Integration with other languages: NumPy integrates seamlessly with

other languages like C and C++, allowing efficient execution of
numerical computations by leveraging optimized code from these
languages.

3. Random number generation: NumPy includes functions for generating

random numbers and random arrays, which are crucial for simulations,
statistical analysis, and machine learning tasks.

4. Linear algebra operations: NumPy provides a wide range of linear

algebra functions, such as matrix multiplication, eigenvalue calculation,
matrix decomposition (e.g., LU, QR), and solving linear systems of
equations.

5. Mathematical functions: NumPy offers a comprehensive set of

mathematical functions for numerical computations. These include
trigonometric functions, exponential and logarithmic functions, statistical
functions, and more.

By utilizing NumPy's efficient array operations and mathematical

functions, developers can perform complex numerical computations with
ease and efficiency. Its integration with other scientific computing
libraries, such as SciPy and Matplotlib, further extends its capabilities
and enables a wide range of scientific and data analysis tasks in Python.
 4.2.3 SCIKIT LEARN

Scikit-learn is a popular machine learning library for Python. It offers a

wide range of algorithms, integrates with other Python libraries like
NumPy and SciPy, and provides a consistent API. It also includes tools
for model selection, evaluation, and performance metrics. Scikit-learn is
robust, well-tested, and leverages Cython for efficient implementations
of certain algorithms.

4.2.4 TENSORFLOW

TensorFlow is a widely used open-source software library that facilitates

numerical computation through data flow graphs. Originally developed
by Google's Machine Intelligence research organization, it was primarily
aimed at conducting research on deep neural networks and machine
learning. However, its flexibility and versatility have made it applicable
in diverse domains. With TensorFlow, users can define mathematical
operations as nodes in a graph, while the edges represent multi-
dimensional data arrays called tensors. This allows for efficient
computation and easy deployment across various devices, including
desktops, mobile devices, and servers, utilizing multiple GPUs and
CPUs. TensorFlow supports multiple platforms such as Windows,
macOS, Linux, iOS, and Android, enabling its adoption in a wide range
of applications beyond neural networks.
 4.2.5 KERAS

Keras is a user-friendly high-level neural networks API that is primarily

implemented in Python and can be utilized on top of frameworks such as
Theano, CNTK, or TensorFlow. It focuses on enabling rapid
experimentation and quick prototyping, allowing researchers to
seamlessly transition from ideas to results. With support for recurrent
networks, convolutional networks, and their combinations, Keras offers
flexibility in modeling various types of neural networks. It is compatible
with both GPU and CPU processing and provides a wide range of pre-
implemented neural network components, including optimizers,
activation functions, layers, and objectives. Additionally, Keras
simplifies working with text and image data by offering convenient tools.
The project is open-source and hosted on GitHub, with active community
support available through forums like GitHub issues, Gitter, and Slack
channels.

4.2.6 COMPILER OPTION

Anaconda is a powerful open-source distribution of the R and Python

programming languages specifically designed for scientific computing,
predictive analytics, and large-scale data processing. Its main objective is
to simplify package management and deployment processes. With
Anaconda, users have access to a wide range of pre-installed packages
and tools commonly used in data analysis and scientific research. The
package management system conda allows for easy management and
version control of software packages, ensuring compatibility and
reproducibility. Anaconda provides a comprehensive platform for data
scientists and researchers to efficiently work with R and Python,
enabling them to focus on their analysis and modeling tasks.
 4.2.7. JUPYTER NOTEBOOK

The Jupyter Notebook is a popular open-source web application that

allows users to create and share interactive documents. It provides a
flexible environment for working with data, enabling tasks such as
data analysis, visualization, transformation, statistical modeling,
machine learning, numerical simulation, data cleaning, and more.

With Jupyter Notebook, users can combine narrative text, live code,
visualizations, and mathematical equations in a single document.
This makes it a versatile tool for data scientists, researchers, and
analysts to explore and communicate their findings effectively.

The interactive nature of Jupyter Notebook fosters collaboration and

makes it easier to present and share data-driven insights.
4.3 Results

Step 1: Dataset-Analysis

Fig 4.1: Stock Dataset Information

Initially, We have performed Data analysis for stock price of companies.
Figure represents the date, open, close, high, low, adjusted close and
volume of stocks details here.

Step 2: Read Dataset

Read Dataset

Post performing data analysis, We had read the dataset and it showed the
dataset information table starting from the tail. There are 4276 data are
available in each companies dataset.
Step 3: Graph of Closing-Price history :

Fig 4.3: Graph of Closing-Price history

Step 4: Pre-processing :-

Fig 4.4: Data Scaling

After Dataset reading, we have performed preprocessing operation on the

dataset. Here we applied Min-Max Scaler to preprocess the dataset. In
preprocessing operation, removed the noise into the data and converted
the data into (0 to 1) form.

Step 5: Train-test Split

After performing preprocessing, we have divided the dataset into training
and testing set. 80% of the data is used for the training while the
remaining 20% of the data is used for testing..
Step 6: Model-fitting of Long Short Term Memory architecture, CNN
architecture.

Long Short Term Memory Summary

Convolutional Neural Network Summary

Step 7: Training Application :

Fig 4.7: Training Process

In the training process, a total of 1548 samples are used for training the
model, while 464 samples are set aside for validation. The data is
processed in batches, with a batch size of 1, meaning that each iteration
of the training algorithm uses a single sample to update the model's
parameters. The training process is performed for one epoch, which
means that the entire training dataset is passed through the model once.
This allows the model to learn from the training data and make
adjustments to its parameters based on the observed outcomes. The
validation samples are used to evaluate the performance of the model
during training and provide feedback on its generalization capabilities.
Step 8: Predicted Result

Predicted Closing-Price

Stock Name
LSTM CNN
GOOGLE 22.5409 8.1499
INFOSYS LTD. 19.4190 16.2015
T-MOBILE 5.4396 6.6478
US Inc.
TESLA 4.7152 3.7276
TWITTER 1.3909 2.5984

Accuracy
Step 9: Predicted Graph

Real vs Predicted value for GOOGLE using LSTM (Plotting)

Real vs Predicted value for GOOGLE using CNN (Plotting)

Real vs Predicted value for INFOSYS LTD. using LSTM (Plotting)

Real vs Predicted value for INFOSYS LTD. using CNN (Plotting)

 Plot for Real vs Predicted value for T-MOBILE-US inc. using
LSTM

Plot for Real vs Predicted value for T-MOBILE US Inc. using CNN
Plot for Real vs Predicted value for TESLA using LSTM

Plot for Real vs Predicted value for TESLA using CNN

Fig 4.17: Plot for Real vs Predicted value for TWITTER using
LSTM

Fig 4.18: Plot for Real vs Predicted value for TWITTER using
CNN
 Chapter Five - Conclusion and Future
Work

In the report, we compare the LSTM model and the CNN model for
predicting the future value of stocks using data from NSE listed
companies. The proposed method demonstrates its ability to identify
interrelationships within the data by utilizing specific information at a
given moment for prediction. While LSTM and CNN models are
commonly used in other time-dependent data analysis, they may not
always capture the rapid changes that occur in the stock market.

Stock market changes do not always follow a regular pattern or a

continuous cycle, and their existence and duration can vary across
companies and sectors. Analyzing such trends and cycles can potentially
yield higher profits for investors. In future work, we aim to include more
stock market data and compare additional models to further improve the
accuracy of predicted stock prices.

To enhance accuracy, future work may involve training the models with
more diverse and detailed data. Additionally, exploring the use of other
algorithms and incorporating them into a hybrid model could lead to
further improvements.
Bibliography & References :

1. J. Du, Q. Liu, K. Chen, and J. Wang, “Forecasting

stock prices in two ways based on LSTM neural
network,” in 2019 IEEE 3rd Information Technology,
Networking, Electronic and Automation Control
Conference (ITNEC), Mar. 2019, pp. 1083–1086, doi:
10.1109/ITNEC.2019.8729026.

2. Z. Yeze and W. Yiying, “Stock Price Prediction

Based on Information Entropy and Artificial Neural
Network,” in 2019 5th International Conference on
Information Management (ICIM), Cambridge, United
Kingdom, Mar. 2019, pp. 248–251, doi:
10.1109/INFOMAN.2019.8714662.

3. “Stock Price Prediction using LSTM, RNN and CNN-

sliding window model - IEEE Conference
Publication.”
https://ieeexplore.ieee.org/document/8126078
(accessed Dec. 27, 2019).

4. Y. Lei, K. Zhou, and Y. Liu, “Multi-Category Events

Driven Stock Price Trends Prediction,” in 2018 5th
IEEE International Conference on Cloud Computing
and Intelligence Systems (CCIS), Nanjing, China,
Nov. 2018, pp. 497–501, doi:
10.1109/CCIS.2018.8691392.

5. S. E. Gao, B. S. Lin, and C.-M. Wang, “Share Price

Trend Prediction Using CRNN with LSTM
Structure,” in 2018 International Symposium on
 Computer, Consumer and Control (IS3C), Dec. 2018,
pp. 10–13, doi: 10.1109/IS3C.2018.00012.

6. M. Usmani, S. H. Adil, K. Raza, and S. S. A. Ali,

“Stock market prediction using machine learning
techniques,” in 2016 3rd International Conference on
computer and Information Sciences (ICCOINS),
2016, pp. 322–327.

7. T. Gao, Y. Chai, and Y. Liu, “Applying long short

term momory neural networks for predicting stock
closing price,” in 2017 8th IEEE International
Conference on Software Engineering and Service
Science (ICSESS), Beijing, China, Nov. 2017, pp.
575–578, doi: 10.1109/ICSESS.2017.8342981.

8. B. Jeevan, E. Naresh, B. P. V. kumar, and P. Kambli,

“Share Price Prediction using Machine Learning
Technique,” in 2018 3rd International Conference on
Circuits, Control, Communication and Computing
(I4C), Bangalore, India, Oct. 2018, pp.
1–4,10.1109/CIMCA.2018.8739647.

9. R. Y. Nivetha and C. Dhaya, “Developing a

Prediction Model for Stock Analysis,” in 2017
International Conference on Technical Advancements
in Computers and Communications (ICTACC),
Melmaurvathur, India, Apr. 2017, pp. 1–3, doi:
10.1109/ICTACC.2017.11.

10. J. Jagwani, M. Gupta, H. Sachdeva, and A. Singhal,

“Stock Price Forecasting Using Data from Yahoo
Finance and Analysing Seasonal and Nonseasonal
Trend,” in 2018 Second International Conference on
 Intelligent Computing and Control Systems
(ICICCS), Madurai, India, Jun.

11. 2018, pp. 462–467, doi:

10.1109/ICCONS.2018.8663035.

12. F. A. Gers, D. Eck, and J. Schmidhuber, “Applying

LSTM to Time Series Predictable through Time-
Window Approaches,” in Artificial Neural Networks
— ICANN 2001, Berlin, Heidelberg, 2001, pp. 669–
676, doi: 10.1007/3-540-44668-0_93.

13. J. Han and C. Moraga, “The Influence of the Sigmoid

Function Parameters on the Speed of
Backpropagation Learning,” in Proceedings of the
International Workshop on Artificial Neural
Networks: From Natural to Artificial Neural
Computation, Berlin, Heidelberg, Jun. 1995, pp. 195–
201, Accessed: Apr. 13, 2022. [Online].

14. J. Brownlee, “Introduction to the Python Deep

Learning Library TensorFlow,” Machine Learning
Mastery, May 04, 2016.
https://machinelearningmastery.com/introduction-
python-deep-learning-library-tensorflow/ (accessed
Apr. 21, 2022).

15. “Python Numpy Tutorial | Learn Numpy Arrays With

Examples,” Edureka, Jul. 14, 2017.
https://www.edureka.co/blog/python-numpy-tutorial/
(accessed Apr. 21).

16. K. Joshi, kartik-joshi/Stock-predection. 2020.

17. K. Team, “Keras documentation: Search Keras
documentation.” https://keras.io/search.html?
query=lstm (accessed Jul. 07, 2022).

18. Y. Kim, “Convolutional Neural Networks for

Sentence Classification,” in Proceedings of the 2014
Conference on Empirical Methods in Natural
Language Processing (EMNLP), Doha, Qatar, Oct.
2014, pp. 1746–1751, doi: 10.3115/v1/D14-1181.

19. B. Athiwaratkun and K. Kang, “Feature

Representation in Convolutional Neural Networks,”
ArXiv150702313 Cs, Jul. 2015, Accessed: Apr. 14,
2022. [Online]. Available:
http://arxiv.org/abs/1507.02313.

20. “Project Jupyter.” https://www.jupyter.org (accessed

Apr. 19, 2022).

21. “Scikit-learn Tutorial: Machine Learning in Python –

Dataquest.” https://www.dataquest.io/blog/sci-kit-
learn-tutorial/ (accessed Apr. 21).

22. “Home - Keras Documentation.” https://keras.io/

(accessed Apr. 21, 2022).

23. “NSE Listed 1000+ Companies’ Historical Data.”

https://kaggle.com/abhishekyana/nse- listed-1384-
companies-data (accessed Jul. 07, 2022)