Unraveling Stock Market Insights with

Advanced Learning Models

A Capstone Project report submitted in partial fulfillment of the

requirements for the Post Graduate Program in Data Science at Praxis Tech
School, Kolkata, India

By

SUYASH MAURYA (Roll No: A23046)

SK MONIRUL ISLAM (Roll No: A23040)
NIKITA DAS (Roll No: A23027)
SK RITURAJ ALAM (Roll No: A23042)
SAYANTAN SAHA (Roll No: A23037)
SHREYAN BHATTACHARJEE (Roll No: A23038)
BAISAKHI BISWAS (Roll No: A23011)
SK MOSIUR RAHMAN (Roll No: A23041)

Of fall 2024 Batch of Data Science

Under the supervision of

Prof. Jaydip Sen

Praxis Tech School, Kolkata, India

1
Acknowledgment

We express our heartfelt gratitude to all those who contributed to the successful completion
of this project. First and foremost, we extend our sincere appreciation to our supervisor Dr.
Jaydip Sen, whose guidance, support, and invaluable insights were instrumental throughout
the research process. Their expertise and encouragement propelled us forward, enabling us to
navigate challenges and achieve our objectives effectively. We are also deeply thankful to
Praxis Tech School, for providing the necessary resources and facilities essential for
conducting this study. The conducive environment fostered a culture of innovation and
collaboration, facilitating our research endeavors. Furthermore, we acknowledge the
participants and stakeholders whose cooperation and contributions were integral to the data
collection process. Their willingness to share insights and provide access to relevant
information significantly enriched our analysis and findings. We are grateful to our
colleagues and peers for their encouragement, constructive feedback, and stimulating
discussions, which enriched our understanding and enhanced the quality of our work. Last
but not least, we extend our heartfelt thanks to our family and friends for their unwavering
support, understanding, and patience throughout this journey. Each of these individuals and
entities played a crucial role in the realization of this project, and for that, we extend our
sincerest appreciation.

Thank you.

2
Abstract
This study endeavors to construct predictive models for forecasting stock prices within the
National Stock Exchange (NSE) of India, focusing on five diverse sectors: IT, steel, auto,
banking, and FMCG. Incorporating data from two major indices, NIFTY and S&P 500. The
analysis encompasses a comprehensive dataset spanning five years, from 2018 to 2022. The
first three years' data (2018-2020) are dedicated to training the models, while the subsequent
two years (2021-2022) are utilized for rigorous testing. Data points are collected at daily
intervals to ensure accuracy and reliability in the analysis process. Advanced deep learning
architectures, including Long-and-Short-Term Memory (LSTM) networks and Convolutional
Neural Networks (CNNs), are employed to extract meaningful insights and patterns from the
data. Both univariate and multivariate analyses are conducted to discern relevant factors
influencing stock price movements across the selected sectors and indices. This research aims
to challenge the efficient market hypothesis by demonstrating the feasibility of accurate stock
price prediction despite market stochasticity. By meticulously designing models and selecting
pertinent variables, the study endeavors to provide stakeholders with actionable insights for
informed decision-making in stock trading and investment activities. The performance of
each predictive model is thoroughly evaluated, considering factors such as prediction
accuracy and computational efficiency. Comparative analyses are conducted to assess the
effectiveness of different model architectures and methodologies.
Ultimately, this research contributes to advancing the field of stock price forecasting within
the NSE of India, offering valuable insights that can enhance risk management strategies and
decision-making processes for market participants.

Keywords: Stock Price Forecasting, Deep Learning, Univariate Analysis, Multivariate

Analysis, Time Series Regression, Root Mean Square Error (RMSE), Long-and-Short-Term
Memory (LSTM) Network, Convolutional Neural Network (CNN)

3
TABLE OF CONTENT
Chapter No. Description Page No
List of Figure 5
1 Introduction 6
2 Related Work 7
3 Methodology 8
4 Deep Learning Models 9
1. Convolutional Neural Network 10
2. Long- and Short-Term Memory Network 13

5 Performance Results and Analysis 16

1. Results of Deep Learning Models
2. Summary of Performance Results of the Models
6 Conclusion and Future Work 30
7 References 31

4
List of Figures

Fig No. Description of Figure Page No

1(a) Variation of RMSE with different days in week for Nifty50 24
(CNN_UNI-5)
1(b) Actual vs Predicted graph for Nifty50 (CNN_UNI-5) 24
2(a) Variation of RMSE with different days in week for Nifty50 25
(CNN_UNI-10)
2(b) Actual vs Predicted graph for Nifty50 (CNN_UNI-10) 25
3(a) Variation of RMSE with different days in week for Nifty50 26
(CNN_MULTI-10)
3(b) Actual vs Predicted graph for Nifty50 (CNN_MULTI-10) 26
4(a) Variation of RMSE with different days in week for S&P500 27
(LSTM_UNI-5)
4(b) Actual vs Predicted graph for S&P500 (LSTM_UNI-5) 27
5(a) Variation of RMSE with different days in week for S&P500 28
(LSTM_UNI-10)
5(b) Actual vs Predicted graph for S&P500 (LSTM_UNI-10) 28
6(a) Variation of RMSE with different days in week for Nifty50 29
(LSTM_MULTI-10)
6(b) Actual vs Predicted graph for Nifty50 (LSTM_MULTI-10) 29

Chapter 1
5
Introduction
The stock market, often referred to as the share market, serves as a primary platform where
business entities and individuals closely monitor their ownership stakes in various companies
and organizations. Given its pivotal role in determining a nation's economic prosperity, the
stock market holds significant sway over economic activities and investor sentiments.
Predicting the trajectory of the stock market proves to be an arduous task owing to the
dynamic and ever-evolving nature of market data. Despite numerous attempts using
conventional methods in the past, accurate stock market predictions have remained elusive,
casting doubt on the reliability of traditional forecasting approaches.

In recent years, the emergence of machine learning (ML) algorithms has brought renewed
hope for improving the accuracy of stock market forecasts. By leveraging sophisticated
algorithms and vast datasets, ML methodologies aim to provide investors with insights into
potential market trends and opportunities for financial growth. However, it's important to
note that while ML algorithms have shown promise in forecasting company stock
performance, they still face challenges in achieving precision.

This study delves into the realm of stock market prediction, shedding light on various
research endeavours aimed at enhancing predictive capabilities. One notable technique
highlighted in the study is the CNN-LSTM Neural Network model, which combines the
strengths of convolutional neural networks (CNNs) and long short-term memory networks
(LSTMs). This innovative approach holds the potential to improve data prediction accuracy
and empower investors with valuable insights for informed decision-making in the realm of
financial investments and strategic planning.

6
 Chapter 2
Related Work
The literature on stock price forecasting is broadly categorized into regression, econometrics,
and machine learning models. Each category offers unique approaches and methodologies to
tackle the complexities of predicting stock market movements.
Regression-based investigations face challenges with high stock market volatility. These
investigations often rely on techniques such as multivariate adaptive regression splines
(MARS) and ordinary least square (OLS) regression. However, the inherent volatility of
stock market data can pose limitations to the accuracy of these models.
Econometric approaches, which utilize time series models like autoregressive integrated
moving average (ARIMA) and Granger Causality, are commonly employed in stock price
forecasting. Yet, when dealing with extremely volatile data, these methods may not always
provide sufficient predictive power.
In recent years, machine learning and deep learning models have emerged as promising
alternatives for stock price prediction. Algorithms such as Convolutional Neural Networks
(CNN) and Long Short-Term Memory (LSTM) networks have demonstrated the ability to
effectively capture complex patterns in stock market data, offering more robust forecasting
capabilities.
Hybrid models, which combine inputs from social web sources with deep learning and
machine learning frameworks, have gained popularity for their potential to improve
prediction accuracy. By blending the strengths of different approaches, hybrid models aim to
overcome the limitations of individual methods and enhance overall forecasting performance.
Deep learning algorithms, in particular, excel in handling the volatility and randomness
inherent in time series data. Their ability to learn intricate patterns and dependencies makes
them well-suited for capturing the dynamic nature of stock market movements, thus offering
valuable insights for investors and financial analysts alike.

7
 Chapter 3
Methodology
The goal of this research is to create accurate prediction models for predicting stock prices on
the Indian National Stock Exchange (NSE). By employing advanced deep learning methods,
we will build strong frameworks that can produce precise stock price forecasts. Furthermore,
our technique incorporates data from a variety of industries and indexes, such as IT, steel,
automotive, banking, FMCG, NIFTY, and S&P 500. With the help of this all-inclusive
method, we can perform in-depth market analysis and improve our comprehension of the
numerous variables affecting stock price movement.

Apart from developing the model, we will meticulously assess its performance by employing
metrics like accuracy and Root mean squared error (MSE). Additionally, we will contrast
different approaches, examining variations in model topologies and training strategies to
maximise computing efficiency and predictive accuracy. Our main objective is to
continuously improve our models in order to increase their efficacy in practical applications.
Yahoo Finance provided daily data for NSE trading days from January 1, 2018, to December
31, 2022. The data covered two indices (NIFTY and S&P 500), five industries (IT, steel,
auto, banking, and FMCG). Variables like date, open, high, low, close, and volume are all
included in the raw data.

The models underwent training from January 01, 2018, to December 31, 2020, historical data,
and testing from January 01, 2021, to December 31, 2022, historical data. The models
produced weekly predictions for the 'Close' value for the test dataset using a multi-step
forecasting technique with walk-forward validation. To enable ongoing improvement and
forecasting for the next week, real 'Close' values were added to the training dataset at the end
of each week.

8
 Chapter 4
Deep Learning Models
Our analysis employs two sophisticated deep learning architectures:
Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM)
network. These models are tailored to predict future stock market trends with
precision and accuracy.

Convolutional Neural Network Models:

CNN_UNI-5: This model adopts a univariate approach, exclusively leveraging
past 'Close' values as input to forecast future 'Close' values, thus capturing
trends based solely on historical closing prices.
CNN_UNI-10: Extending the temporal scope, CNN_UNI-10 utilizes a longer
historical context (10 days) of past 'Close' values, enabling deeper insight into
evolving market trends while maintaining the univariate focus.
CNN_MULTI-10: In contrast to univariate models, CNN_MULTI-10 embraces
a multivariate perspective by incorporating all five variables (Open, High, Low,
Close, Volume) as input. By considering a broader range of market indicators,
this model provides a comprehensive outlook on future 'Close' values.

Long and Short-Term Memory Network Models:

LSTM_UNI-5: Employing a univariate strategy akin to CNN_UNI-5, this
model capitalizes on past 'Close' values to predict future 'Close' values, thereby
capturing nuanced temporal patterns in isolation.
LSTM_UNI-10: Analogous to LSTM_UNI-5, LSTM_UNI-10 extends its
predictive horizon by incorporating a longer historical context (10 days) of past
'Close' values, enriching its forecasting capabilities.
LSTM_MULTI-10: Departing from the univariate paradigm, LSTM_MULTI-
10 embraces a multivariate approach by integrating all five variables (Open,
High, Low, Close, Volume) as input. This comprehensive model leverages a
holistic view of market dynamics, enabling nuanced predictions of future 'Close'
values.

9
Convolutional Neural Network Models

1. CNN_UNI-5

The univariate model processes one-week (N = 5) preceding data, reading five days in
three time-steps with a single convolution layer featuring 16 maps and a kernel size of
three, extracting 16 features per reading. It utilizes max pooling of size two to reduce
feature map size before flattening into a single vector. A fully-connected layer
precedes the output layer, utilizing ReLU activation and ADAM optimizer, except for
the final layer which uses sigmoid activation. Training involves a batch size of 4, 20
epochs, and MSE loss for RMSE computation.

Architecture for CNN_UNI-5 model

10
2. CNN_UNI-10

The univariate model processes data from the last two weeks (N = 10) as input. The
sequence of 10 days is read in three time-steps, repeated 16 times with a single
convolution layer featuring 16 maps and a kernel size of 3. Max pooling of size 2
reduces the feature map size before flattening into a single vector of size 64. A fully-
connected layer interprets the flattened vector before the output layer predicts the
open values. ReLU activation and ADAM optimizer are utilized, except for the final
layer which employs sigmoid activation.

Architecture for UNI_CNN-10 model

11
3. CNN_MULTI-10

For multivariate analysis, the model utilizes data from the previous two weeks (N =
10) with all five variables. The sequence of 10 days is read in three time-steps,
repeated 16 times using a single convolution layer featuring 16 feature maps and a
kernel size of 3. Max pooling of size 2 reduces feature map size before flattening into
a single vector of size 64. A fully connected layer interprets the flattened vector
before the output layer predicts open values. ReLU activation and ADAM optimizer
are used, except for the final layer which employs sigmoid activation. Training
involves a batch size of 16, 70 epochs, MSE loss, and computation of RMSE values.

Architecture for CNN_MULTI-10 model

12
Long and Short-Term Memory Models

1. LSTM_UNI-5

The univariate model utilizes data from the past week (N = 5) as input. The LSTM layer
comprises 200 nodes, processing the input sequence of shape (5,1) to generate a 200-
element vector, representing features extracted from the input values. Subsequently, the
first dense layer consists of 100 nodes, receiving the 200-element vector from the
preceding LSTM layer. The output layer, with nodes equal to the number of outputs,
employs ReLU activation. MSE serves as the loss function and ADAM as the optimizer
in the output layer. Training encompasses 20 epochs with a batch size of 4.

Architecture for UNI_LSTM-5 model

13
2. LSTM_UNI-10
The univariate model utilizes past week's data (N = 10) as input. The LSTM layer
consists of 200 nodes, processing the input sequence of shape (10,1) to generate a
200-element vector representing extracted features. The first dense layer comprises
100 nodes, receiving the 200-element vector from the preceding LSTM layer. The
output layer, with nodes equal to the number of outputs, employs ReLU activation.
MSE serves as the loss function and ADAM as the optimizer in the output layer.
Training encompasses 20 epochs with a batch size of 4.

Architecture for LSTM_UNI-10 model

14
3. LSTM_MULTI-10
The multivariate model utilizes past week's data (N = 10) as input. The LSTM layer
consists of 200 nodes, processing the input sequence of shape (10,5) to generate a
200-element vector representing extracted features. The first dense layer comprises
100 nodes, receiving the 200-element vector from the preceding LSTM layer. The
output layer, with nodes equal to the number of outputs, employs ReLU activation.
MSE serves as the loss function and ADAM as the optimizer in the output layer.
Training encompasses 70 epochs with a batch size of 16.

Architecture for LSTM_MULTI-10 model

15
 Chapter 5
Performance Results and Analysis
Result of Deep Learning Models

CNN_UNI-5

Data Time RMSE/Mean Mon Tue Wed Thu Fri RMSE

Tata Steel 16.5 8.9 5.3 5.2 5.2 6.5 6.6 5.8

Tata Consultancy
17.4 3.9 67.4 82.3 86.3 107.9 115.1 93.4
Services

State Bank of India 21.6 6.6 16.0 19.7 23.5 25.1 25.5 22.3

NIFTY 21.3 2.8 313.5 333.5 359.6 403.7 452.0 375.8

S&P 16.4 3.2 91.0 100.4 102.8 116.3 134.0 109.9

Maruti Suzuki India

21.9 4.0 250.6 267.6 285.8 300.2 339.8 290.4
Limited

Mahindra & Mahindra

22.1 5.2 33.0 36.5 37.9 41.1 45.7 39.1
Limited

JSW Steel Limited 21.4 7.1 22.5 25.1 27.1 32.7 34.6 28.8

Infosys Limited 21.3 5.2 43.2 49.0 50.1 55.6 62.2 52.4

ITC Limited 16.6 3.4 5.9 6.2 7.2 8.2 9.2 7.5

Hindustan Unilever
20.3 3.8 62.3 67.6 71.1 85.7 88.0 75.6
Ltd

HDFC Bank Limited 19.9 4.0 43.2 45.8 47.5 50.6 53.7 48.3

MEAN 19.7 4.917 79.5 86.6 92.0 102.8 113.9 95.8

MIN 16.4 2.8 5.3 5.2 5.2 6.5 6.6 5.8

MAX 22.1 8.9 313.5 333.5 359.6 403.7 452.0 375.8

SD 2.3 1.8 98.9 105.0 112.8 123.6 139.5 116.8

16
CNN_UNI-10

Data Time RMSE/Mean Mon Tue Wed Thu Fri RMSE

Tata Steel 18.5 8.6 5.1 4.8 5.1 6.4 6.6 5.6

Tata Consultancy 131.

18.1 4.7 98.7 92.4 101.4 129.9 111.9
Services 0

State Bank of India 21.4 6.3 16.5 18.6 18.9 23.1 26.6 21.0

355. 535.
NIFTY 17.6 3.4 489.2 397.6 482.4 456.8
8 4

117. 129.
S&P 17.4 3.8 109.5 134.9 154.2 130.0
4 5

Maruti Suzuki India 287. 289.

19.0 4.0 273.1 278.5 331.2 292.7
Limited 7 4

Mahindra & Mahindra

21.6 5.2 34.6 33.8 36.9 39.3 46.6 38.5
Limited

JSW Steel Limited 18.0 6.8 21.9 27.5 24.3 30.1 32.4 27.5

Infosys Limited 21.1 5.5 47.4 51.4 56.8 57.4 62.3 55.3

ITC Limited 16.8 4.6 5.7 8.6 9.5 11.8 13.0 10.1

Hindustan Unilever 100.

20.5 4.5 58.8 79.7 89.2 107.1 88.6
Ltd 1

HDFC Bank Limited 21.7 4.0 43.7 48.5 45.2 53.9 52.1 48.8

117.
MEAN 19.3 5.1 100.3 93.9 99.9 120.4 107.2
3

MIN 16.8 3.4 5.1 4.8 5.1 6.4 6.6 5.6

355. 535.
MAX 21.7 8.6 489.2 397.6 482.4 456.8
8 4

112. 153.
SD 1.8 1.5 142.8 120.6 145.0 135.5
7 3

17
 Data Tim RMSE/ Mon Tue Wed Thu Fri RMSE
e Mean

Tata Steel 18.0 142.3 89.0 100.8 88.4 97.5 89.6 93.2

Tata 20.9 98.9 2356.5 2998.7 2306.5 2130.1 1961.3 2377.0

Consultan
cy Services

State Bank 19.2 2317.1 4712.4 11074.2 8480.5 5144.4 7713.6 7781.1
of India

NIFTY 20.8 3.7 409.8 430.0 499.0 491.2 606.8 492.2

S&P 21.3 14986.3 176982. 397882. 468818. 335330. 902040. 516826.

9 8 3 5 7 5

Maruti 19.2 8.6 681.3 599.0 635.1 569.0 627.6 623.5

Suzuki
India
Limited

Mahindra 23.1 115.4 972.4 649.6 656.2 990.4 970.8 862.7

&
Mahindra
Limited

JSW Steel 18.7 838.0 2626.8 5036.4 2127.1 3765.6 2556.3 3391.3
Limited

Infosys 21.7 126.8 1065.4 1330.5 1466.3 1208.7 1231.4 1267.5

Limited

ITC 33.5 3221.4 8441.3 4249.4 12144.6 1338.4 2183.0 6976.6

Limited

Hindustan 22.6 95.0 1685.9 2305.4 1942.3 1732.9 1626.6 1875.0

Unilever
Ltd

HDFC 19.3 92.0 1108.9 1100.2 1058.7 1196.5 1138.1 1121.4

Bank
Limited

MEAN 21.5 1837.1 16761.1 35646.4 41685.3 29499.6 76895.5 45307.3

MIN 18.0 3.7 89.0 100.8 88.4 97.5 89.6 93.2

MAX 33.5 14986.3 176982. 397882. 468818. 335330. 902040. 516826.

9 8 3 5 7 5

SD 4.1 4270.5 50510.5 114116. 134561. 96322.5 259860. 148511.

0 9 9 0

CNN_MULTI-10

18
LSTM_UNI-5

Data Time RMSE/Mean Mon Tue Wed Thu Fri RMSE

Tata Steel 32 8.8 4.6 5.2 4.8 6.8 7 5.8

Tata Consultancy 32.5 4.2 62.9 111.7 101.4 104.7 120.3 102.1
Services

State Bank of India 33.2 5.9 15.8 16 19.7 20.7 25.6 19.9

NIFTY 32.8 3.7 309.6 491.3 354.3 524.8 693.9 493.9

S&P 32.5 3 74.7 85.2 119 107.3 119.9 102.9

Maruti Suzuki India 34.8 3.8 271.2 237.7 242.1 268.6 332 272.4
Limited

Mahindra & Mahindra 32.5 6 33.6 45.6 50.3 46.8 47.7 45.2
Limited

JSW Steel Limited 32.7 7.6 24.8 21.9 37.5 33.7 33 30.7

Infosys Limited 34.4 4.7 37.9 42.4 44.1 52 57.1 47.2

ITC Limited 34.4 3.9 7.9 7.2 8.2 8.7 9.6 8.4

Hindustan Unilever 32.8 3.4 49.7 54.1 68.2 75.1 80.5 66.6
Ltd

HDFC Bank Limited 33.8 3.5 36 40.2 42.8 45.3 48.9 42.8

MEAN 33.2 4.97 77.4 96.5 91 107.9 131.3 103.2

MIN 32 3 4.6 5.2 4.8 6.8 7 5.8

MAX 34.8 8.8 309.6 491.3 354.3 524.8 693.9 493.9

SD 0.9 1.8 102 139.8 105.3 149.1 197.7 142.5

19
LSTM_UNI-10

Data Time RMSE/Mean Mon Tue Wed Thu Fri RMSE

Tata Steel 94.3 10.8 7.3 5.4 8.2 6.6 7.6 7.1

Tata Consultancy 92.9 6.4 147.1 167.9 129.5 149.8 168.1 153.2
Services

State Bank of India 95.8 6.5 17.3 19.2 25.1 21.6 25.3 21.9

NIFTY 94.4 3.8 510.2 461.7 479.7 515.3 539.4 502.0

S&P 68.0 3.2 89.1 94.2 100.0 130.8 128.9 110.0

Maruti Suzuki India 93.1 4.6 315.6 266.9 301.6 357.7 393.5 330.0
Limited

Mahindra & Mahindra 94.0 7.0 46.0 48.1 46.9 57.1 60.4 52.0
Limited

JSW Steel Limited 95.8 10.7 37.6 39.6 42.4 46.4 49.1 43.2

Infosys Limited 65.8 5.7 49.4 48.2 60.3 59.8 66.1 57.2

ITC Limited 64.4 5.3 9.6 11.9 11.8 13.1 11.3 11.6

Hindustan Unilever 93.9 4.1 60.3 76.1 76.9 82.5 98.9 79.9
Ltd

HDFC Bank Limited 69.9 5.0 49.8 61.4 49.9 59.4 78.2 60.6

MEAN 85.2 6.1 111.6 108.4 111.0 125.0 135.6 119.1

MIN 64.4 3.2 7.3 5.4 8.2 6.6 7.6 7.1

MAX 95.8 10.8 510.2 461.7 479.7 515.3 539.4 502.0

SD 13.5 2.5 151.5 133.8 140.4 155.4 164.4 149.2

20
LSTM_MULTI-10

Data Tim RMSE/ Mon Tue Wed Thu Fri RMSE

e Mean

Tata Steel 116. 58807.7 52945.8 30668.1 41098.4 34895.8 27774.6 38528.0
8

Tata 72.6 210.0 3566.4 6526.8 2608.9 5731.5 5690.7 5047.0

Consultan
cy
Services

State 74.0 5060.7 18689.4 10883.4 16358.4 23567.1 12378.4 16994.0

Bank of
India

NIFTY 67.5 6.8 806.0 484.3 1228.4 1003.8 814.9 901.7

S&P 92.5 40555.1 1059661 1359717 1076852 2071776 1164894 1398610

.0 .1 .0 .1 .7 .6

Maruti 92.3 13.5 931.0 859.1 976.5 1212.5 850.8 974.9

Suzuki
India
Limited

Mahindra 92.9 1018.0 10218.6 9274.6 7051.9 2704.2 6471.7 7607.4

&
Mahindra
Limited

JSW Steel 71.5 4799.4 13329.1 27478.9 14430.0 22425.1 15565.8 19422.5
Limited

Infosys 68.8 1120.3 9036.5 3496.0 12383.7 17070.5 9395.9 11197.5

Limited

ITC 71.0 3644.3 12253.2 9595.8 4265.3 4531.6 5522.0 7892.3

Limited

Hindusta 75.3 61.4 1060.7 944.5 1023.1 1205.1 1679.1 1211.3

n
Unilever
Ltd

HDFC 93.0 189.3 2804.0 2748.9 2270.8 1447.9 1992.5 2308.3

Bank
Limited

MEAN 82.3 9623.9 98775.1 121889. 98379.0 182297. 104419. 125891.

8 6 3 3

21
MIN 67.5 6.8 806.0 484.3 976.5 1003.8 814.9 901.7

MAX 116. 58807.7 1059661 1359717 1076852 2071776 1164894 1398610

8 .0 .1 .0 .1 .7 .6

SD 15.0 19202.1 302939. 389941. 308349. 595139. 334053. 400948.

0 9 1 5 2 7

22
Summary of Performance Results of the
Models

We observe that CNN -2 model – a univariate CNN model with past two week’s data as its
input – turns out to be the fastest model in execution.
The model CNN -1 – a univariate CNN model with past one week’s data as its input – turned
out to be the most accurate model.

Execution Time RMSE/MEAN

Rank Model Value Rank Model Value

1 CNN_UNI-10 19.3 1 CNN_UNI-5 4.91

2 CNN_UNI-5 19.7 2 LSTM_UNI-5 4.97

3 CNN_MULTI-10 21.5 3 CNN_UNI-10 5.1

4 LSTM_UNI-5 33.2 4 LSTM_UNI-10 6.1

5 LSTM_MULTI-10 82.3 5 CNN_MULTI-10 1837.1

6 LSTM_UNI-10 85.2 6 LSTM_MULTI-10 9623.9

23
List of Figures

Fig 1(a) Variation of RMSE with different days in week for Nifty50 (CNN_UNI-5)

Fig 1(b) Actual vs Predicted graph for Nifty50 (CNN_UNI-5)

24
Fig 2(a) Variation of RMSE with different days in week for Nifty50 (CNN_UNI-10)

Fig 2(b) Actual vs Predicted graph for Nifty50 (CNN_UNI-10)

25
Fig 3(a) Variation of RMSE with different days in week for Nifty50 (CNN_MULTI-10)

Fig 3(a) Actual vs Predicted graph for Nifty50 (CNN_MULTI-10)

26
Fig 4(a) Variation of RMSE with different days in week for S&P500 (LSTM_UNI-5)

Fig 4(b) Actual vs Predicted graph for S&P500 (LSTM_UNI-5)

27
Fig 5(a) Variation of RMSE with different days in week for S&P500 (LSTM_UNI-10)

Fig 5(b)Actual vs Predicted graph for S&P500 (LSTM_UNI-10)

28
Fig 6(a) Variation of RMSE with different days in week for Nifty50 (LSTM_MULTI-10)

Fig 6(b) Actual vs Predicted graph for Nifty50 (LSTM_MULTI-10)

29
 CHAPTER 6

Conclusion and Future Work

Our study focused on developing deep learning-driven regression models for forecasting
daily stock prices on the National Stock Exchange (NSE) of India. Six models were
constructed and tested, including two univariate CNN models, two univariate LSTM models,
and one multivariate model combining CNN and LSTM. Using data from January 1, 2021, to
December 31, 2022, we achieved exceptional accuracy levels across all models. Particularly,
the CNN model utilizing data from the previous one and two weeks emerged as the top
performer in terms of accuracy and speed.

Looking ahead, we plan to explore the integration of generative adversarial networks (GANs)
to potentially enhance prediction accuracy. Additionally, we aim to investigate the
incorporation of alternative data sources, such as sentiment analysis from news articles or
social media data, to improve prediction robustness. Furthermore, we will consider the
inclusion of exogenous variables, such as macroeconomic indicators, to refine forecasting
models. Additionally, we aim to explore various hybrid LSTM models and integrate LSTM,
CNN, and GAN models to further advance stock price prediction research.

Through these future directions, we seek to contribute to the development of more accurate
and reliable stock price forecasting models, aiding decision-making in financial markets.

30
 Chapter 7

References
Damrongsakmethee,T. and Neagoe,V.-E. "Stock Market Prediction Using a Deep
Learning Approach," 2020 12th International Conference on Electronics,
Computers and Artificial Intelligence (ECAI), Bucharest, Romania, 2020, pp.
1-6, DOI: 10.1109/ECAI50035.2020.9223142.

Mehtab, S. and Sen, J. "Stock Price Prediction Using CNN and LSTM-Based Deep
Learning Models," 2020 International Conference on Decision Aid Sciences
and Application (DASA), Sakheer, Bahrain, 2020, pp. 447-453, doi:
10.1109/DASA51403.2020.9317207.

Maiti, A. and Shetty D, P. "Indian Stock Market Prediction using Deep Learning,"
2020 IEEE Region 10 Conference (TENCON), Osaka, Japan, 2020, pp. 1215-
1220, doi: 10.1109/TENCON50793.2020.9293712.

P. S and V. P. R, "Stock Price Prediction using Machine Learning and Deep

Learning," 2021 IEEE Mysore Sub Section International Conference
(MysuruCon), Hassan, India, 2021, pp. 660-664,
DOI:10.1109/MysuruCon52639.2021.9641664.

P. Praveen, S. Shravani, R. Srija and M. Tajuddin, "A Model to Stock Price

Prediction using Deep Learning," 2023 International Conference on
Sustainable Computing and Smart Systems (ICSCSS), Coimbatore, India,
2023, pp. 242-252, doi: 10.1109/ICSCSS57650.2023.10169558.

Biswas, M., Shome, A., Islam, M. A., Nova, A. J. and Ahmed, S. "Predicting Stock
Market Price: A Logical Strategy using Deep Learning," 2021 IEEE 11th
IEEE Symposium on Computer Applications &Industrial Electronics
(ISCAIE), Penang,Malaysia, 2021, pp.218-223,
doi:10.1109/ISCAIE51753.2021.9431817

Hossain, M. A., Karim, R., Thulasiram, R., Bruce, N. D. B. and Y. Wang, "Hybrid

31
 Deep Learning Model for Stock Price Prediction," 2018 IEEE Symposium
Series on Computational Intelligence (SSCI), Bangalore, India, 2018, pp.
1837-1844, doi: 10.1109/SSCI.2018.8628641.

Khan, S., Rabbani, M. R., Bashar, A. and Kamal, M., "Stock Price Forecasting
Using Deep Learning Model," 2021 International Conference on Decision Aid
Sciences and Application (DASA), Sakheer, Bahrain, 2021, pp. 215-219, doi:
10.1109/DASA53625.2021.9682319.

Parida, N., Balabantaray, B. K., Nayak, R. and Rout, J. K., "A Deep Learning based
Approach to Stock Market Price Prediction using Technical indicators," 2023
International Conference on Advances in Intelligent Computing and
Applications (AICAPS), Kochi, India, 2023, pp. 1-7, doi:
10.1109/AICAPS57044.2023.10074445.

Akita, R., Yoshihara, A., Matsubara, T., and Uehara, K., "Deep learning for stock
prediction using numerical and textual information," 2016 IEEE/ACIS 15th
International Conference on Computer and Information Science (ICIS),
Okayama, Japan, 2016, pp. 1-6, doi: 10.1109/ICIS.2016.7550882

Singh, T., Kalra, R., Mishra, S. et al. "An efficient real-time stock prediction
exploiting incremental learning and deep learning". Evolving Systems 14,
919–937 (2023). https://doi.org/10.1007/s12530-022-09481-x.

Li, Y., Pan, Y. "A novel ensemble deep learning model for stock prediction based on
stock prices and news". Int J Data Sci Anal 13, 139–149 (2022).
https://doi.org/10.1007/s41060-021-00279-9.

Yu, P., Yan, X., "Stock price prediction based on deep neural networks". Neural
Comput & Applic 32, 1609– 1628 (2020). https://doi.org/10.1007/s00521-019-
04212-x.

Billah, M.M., Sultana, A., Bhuiyan, F. et al., "Stock price prediction: comparison of
different moving average techniques using deep learning model". Neural
Comput & Applic (2024). https://doi.org/10.1007/s00521- 023-09369-0.

Shah, A., Gor, M., Sagar, M. et al., "A stock market trading framework based on

32
 deep learning architectures". Multimed Tools Appl 81, 14153–14171 (2022).
https://doi.org/10.1007/s11042-022-12328-x.

Ding, G., Qin, L., "Study on the prediction of stock price based on the associated
network model of LSTM". Int. J. Mach. Learn. & Cyber. 11, 1307–1317
(2020). https://doi.org/10.1007/s13042-019-01041-1.

Bandhu, K.C., Litoriya, R., Jain, A. et al., "An improved technique for stock price
prediction on real-time exploiting stream processing and deep learning".
Multimed Tools Appl (2023). https://doi.org/10.1007/s11042-023-17130-x.

Swathi, T., Kasiviswanath, N. & Rao, A.A. "An optimal deep learning-based LSTM
for stock price prediction using twitter sentiment analysis". Appl Intell 52,
13675–13688 (2022). https://doi.org/10.1007/s10489-022- 03175-2.

Olorunnimbe, K., Viktor, H. "Deep learning in the stock market—a systematic

survey of practice, backtesting, and applications". Artif Intell Rev 56, 2057–
2109 (2023). https://doi.org/10.1007/s10462-022- 10226-0.

Mehtab S. and Sen, J. (2022) “Analysis and forecasting of financial time series using
CNN and LSTM-based deep learning models,” in: Sahoo, J. P., Tripathy, A.
K., Mohanty, M., Li, K. C., Nayak, A, K. (eds), Advances in Distributed
Computing and Machine Learning, Lecture Notes in Networks and Systems,
Vol 302, pp. 405-423, Springer, Singapore, doi: doi.org/10.1007/978-981-16-
4807-6_39.

33