Table of Contents

ACKNOWLEDGEMENT..........................................................................................
Declaration..............................................................................................................
Chapter 1 Introduction...........................................................................................
Purpose................................................................................................................
Scope...................................................................................................................
Stock analysis: fundamental analysis vs. technical analysis...............................
Problem Statement..............................................................................................
Chapter 2 Literature Survey.................................................................................
2.1 Introduction to Literature Review 7
Chapter 3 Methodology.........................................................................................14
3.1 PROPOSED SYSTEMS 14
3.1.1 Time Series Analysis 14
3.1.2 Long short-term memory network: 20
3.2 SYSTEM ARCHITECTURE 25
Chapter 4 Design..........................................................................................26
4.1 Structure Chart 26
4.2 UML Diagrams 27
4.2.1 Use Case Diagram 29
4.2.2 Sequence Diagram 30
4.2.3 Activity Diagram 31
4.2.4 Collaboration Diagram 32
4.2.5 Flow Chart 33
4.2.6 Component Diagram 34
Chapter 5 Experiment Analysis...........................................................................35
5.1 System configuration 35
5.2 Sample code 36
Chapter 6 Conclusion And Future Work.................................................66
6.1 Conclusion 66
6.2 Future work 66

ACKNOWLEDGEMENT
I would like to thanks my project guide Indra Nath Sahu sir, for
contributing a lot of efforts in providing me with resources, study
materials and reading manuscript and for devoting his valuable time
and always begins a constant source of advice and inspiration
without which this work wouldn’t have been possible.
I would like to thanks Dr. A.K. Mahto, coordinator of MCA/MSC
IT, all the faculties and other staff of department MASTER IN
COMPUTER APPLICATION of DSPMU, for their cooperation
and encouragement throughout the period of my projects and stay in
DSPMU, Ranchi.

Date: Sawan
Kumar

Declaration
I, Sawan Kumar hereby declare that the project report “Stock
Market Price Prediction “is based on my own work carried out
during the course of our study under the supervision of Pooja Jha
Mam, Department of MCA, DSPMU RANCHI.
I assert the statements made and conclusions drawn are an outcome
of my research work. I further certify that

I. The work contained in the report is original and has been done
by me under the general supervision of my supervisor.
II. The work has not been submitted to any other Institution for
any other degree/diploma/certificate in this university or any
other University of India or abroad.
III. We have followed the guidelines provided by the university in
writing the report.
IV. Whenever we have used materials (data, theoretical analysis, and
text) from other sources, we have given due credit to them in the text
of the report and giving their details in the references.
Chapter 1 Introduction

Purpose
Stock Market Price is known for being volatile, dynamic and nonlinear.
Accurate stock price prediction is extremely challenging because of
multiple (macro and micro) factors, such as politics, global economic
conditions, unexpected events, a company’s financial performance,
and so on. 
But, all of this also means that there’s a lot of data to find patterns in.
So, financial analysts, researchers, and data scientists keep exploring
analytics techniques to detect stock market trends. This gave rise to the
concept of algorithmic trading, which uses automated, pre-
programmed trading strategies to execute orders.
Scope
Despite the volatility, stock prices aren’t just randomly generated
numbers. So, they can be analysed as a sequence of discrete-time data;
in other words, time-series observations taken at successive points in
time (usually on a daily basis). Time series forecasting (predicting
future values based on historical values) applies well to stock
forecasting.
Because of the sequential nature of time-series data, we need a way to
aggregate this sequence of information. From all the potential
techniques, the most intuitive one is MA with the ability to smooth out
short-term fluctuations.

Stock analysis: fundamental analysis vs. technical analysis

When it comes to stocks, fundamental and technical analyses are at
opposite ends of the market analysis spectrum.

 Fundamental analysis:

5
 Evaluates a company’s stock by examining its intrinsic
o
value, including but not limited to tangible assets,
financial statements, management effectiveness, strategic
initiatives, and consumer behaviors; essentially all the
basics of a company.
o Being a relevant indicator for long-term investment, the
fundamental analysis relies on both historical and
present data to measure revenues, assets, costs,
liabilities, and so on.
o Generally speaking, the results from fundamental
analysis don’t change with short-term news. 
 Technical analysis:
o Analyzes measurable data from stock market activities,
such as stock prices, historical returns, and volume of
historical trades; i.e. quantitative information that could
identify trading signals and capture the movement
patterns of the stock market. 
o Technical analysis focuses on historical data and current
data just like fundamental analysis, but it’s mainly used
for short-term trading purposes.
o Due to its short-term nature, technical analysis results
are easily influenced by news.
o Popular technical analysis methodologies include moving
average (MA), support and resistance levels, as well
as trend lines and channels. 

For our exercise, we’ll be looking at technical analysis solely and

focusing on the Simple MA and Exponential MA techniques to predict
stock prices. Additionally, we’ll utilize LSTM (Long Short-Term
Memory), a deep learning framework for time-series, to build a
predictive model and compare its performance against our technical
analysis. 
As stated in the disclaimer, stock trading strategy is not in the scope of
this article. I’ll be using trading/investment terms only to help you
better understand the analysis, but this is not financial advice. We’ll be
using terms like:

 trend indicators: statistics that represent the trend of stock prices,

6
  medium-term movements: the 50-day movement trend of stock
prices.

Problem Statement
Time Series forecasting & modelling plays an
important role in data analysis. Time series analysis is a
specialized branch of statistics used extensively in fields such as
Econometrics & Operation Research. Time Series is being widely
used in analytics & data science. Stock prices are volatile in
nature and price depends on various factors. The main aim of
this project is to predict stock prices using Long short-term
memory (LSTM).

Chapter 2 Literature Survey

In this section we will review Key Terms, Technical Indicators and

probabilistic models in depth

2.1 Introduction to Literature Review

Stock market prediction is an area of research which walks hand in
hand with all the other businesses in the world. Hence, it is known as
'Mother of all Businesses’. Before trying to jump into analysis we need
to understand the following terms.

2.1.1 Key Terms

2.1.1.1 Interval
Millions of trades take place every second. Hence, for analysis we need
to classify these trades based on the interval at which they took place.
We can divide these trades into intra-day and long-term intervals.
7
Intra-day can be sub-classified into 1 minute, 2, 5, 10, 15, 30 and 60
minutes. Long term intervals can be classified into daily, weekly,
monthly and so on.

2.1.1.2 Tick Prices

At every interval any stock will have 4 types of prices associated
with it. High price, Low price, Open Price and Close price. High price is
the highest value at which it was traded in that particular interval. Low
price is the lowest price which it reached in that interval. Open price is
the first trade which took place at that particular interval and Close
price the price at which the stock was last traded at that interval

2.1.1.3 Trend
At any instance of time, any stock will have either higher demand than
supply or lower demand than available supply. Hence, we can classify
the Trend into two types namely Bearish and Bullish. A stock is said to
be in Bullish Trend if it has higher demand than its supply at that
instance of time. If the stock has higher available supply when
compared with demand, it is said it be in Bearish Trend.

2.1.2 Data Representation

Data need to be represented in some format before we can start
analyzing it. In this section we will discuss how can we represent data
for better interpretation.

2.1.2.1 Japanese Candle

8
 Figure 1 Japanese Candle
Figure 1 represents two different candle sticks. The green candle
represents gain, and red candle represents loss. This candle stick
always contains all attributes associated with an interval (Example:
tick prices). Since it represents an interval what we are looking at, we
will refer to a candle stick as an interval in the rest of this report.

2.1.2.2 Line Graph

Line graph is a graphical way of representing data points in stock
market based only on close price or current trading price of the stock.

Figure 2. Line Graph

9
2.1.3 Accuracy
Accuracy is defined as the result of number of right decisions made
divided by total number of decisions.
2.1.3.1 Accuracy based on Close Price
When we calculated accuracy, if we relied on next interval's close
price for decision verification, we have mentioned it as Accuracy based
on Close price in the rest of the report. When we make a Buy call, to
verify based on close price, we check if next interval's close price is
above current interval's close price. If so, we consider the decision as a
right decision. Also, if we make a Short sell, to verify the decision based
on close price, we check if next interval's close price is lesser than that
of current interval's close price. If so, we consider the decision as right
decision.

2.1.3.2 Accuracy based on Low/High Price

When calculating accuracy, if relied on next interval's low/high price
for decision verification, we have mentioned it as Accuracy based on
high/low price in the rest of the report. When we make a Buy call, to
verify the decision based on high price, we check if current interval's
close price is lesser than next interval's high price. [4] If so, we
consider the decision as a right decision. Similarly, if we make a Short
sell, to verify the decision based on low price, we check if next
interval's low price is lesser than current interval's close price. If so, we
consider this decision as right decision.

2.1.3.3 Short Sell

Short sell is a type of trade where we sell the stock first, before buying it
and buy it at a later point of time. Investors perform this trade if they think
stock market is going to follow Bearish trend.

2.1.3.4 Buy Call

Buy call is a type of trade where investors invest by buying the
stock first and selling it at a later point of time. They perform this trade
if they think, the stock in which they are investing will follow Bullish
trend and can yield them profit.

10
2.1.4 Technical Indicators
Technical Indicators are the properties associated with any stock
based on its tick prices (2.1.1.2). Calculation of each indicator is
mentioned in the following subsections.
Interpretation of each of the indicator is explained in detail in section
3.0.

2.1.4.1 Simple Moving Average (SMA)

Simple Moving Average is calculated exclusively based on close price
of the stock which we are trying to analyze. For instance, if we need to
calculate SMA of 'x' intervals we need to get close prices of previous 20
intervals and divide it by 'x'. Hence the first available SMA value will
correspond to xth interval. To calculate we add all close prices starting
from the current interval looking back for n number of intervals. In
this case n stands for number of intervals for which we need to
calculate SMA.

2.1.4.2 EXPONENTIAL MOVING AVERAGE (EMA)

Exponential Moving Average is also calculated based on close price of
the stock which we are analyzing. If we need to calculate EMA of 'x'
intervals, we first need to calculate SMA (2.1.2.1) till xth interval and for
every subsequent interval we calculate EMA based on the following
formula.

EMA =

Hence, the first available EMA value will be corresponding to xth

interval. EMA
moves hand in hand with price of stock. Higher the number of
intervals we choose to calculate EMA, higher the stability of EMA. [18]
Hence, we chose 2, 5, 10, 20, 50 and 100 intervals for calculating EMA.
2.1.4.3 Relative Strength Index (RSI)
Relative Strength Index is calculated based on SMA and close price of
11
the stock for the given interval. We must get familiar with the following
terms to better understand the calculation of RSI: Gain, Loss. If the
close price of a stock at a given interval is greater than its open price,
the stock resulted in Gain, vice versa it resulted in Loss. [18]
Here are the formulae to calculate RS and RSI. [3, 4]

RS = Average Gain/ Average Loss

RSI indicates the strength of the current trend. Higher the value
of interval we choose, we get stable RSI values. We need to find out a
threshold value. If RSI falls below its threshold, it is an indication of
sellers taking over buyers. If RSI value rises over its threshold, it
indicates that buyers are taking over sellers and stock prices will go
high. [14]
2.1.4.4 Bollinger Band (BB)
Bollinger Band is calculated based on Standard Deviation
(2.1.2.3) and close price of the stock at a given interval. [12] Bollinger
Bands are calculated based on the following formulae:

Bollinger Band’s values provide an insight on how much more the

stock can rise if it is in Bullish trend. Or, how much can it fall, if the

stock is in Bearish trend.

2.1.4.5 Fast Stochastic (FS)

Fast Stochastic is the technical indicator which involves two
values, namely, %k and %d. For a given interval these values are
12
calculated based on current close price of the stock,[18] lowest low
price in the look-back period and highest high price of the stock in the
look-back period. Lowest low and highest high are the lowest price and
highest price at which the stock was traded in the given look-back
period with respect to given interval respectively.
Here are the formulae to calculate %k and %d:

At a given interval %k’s value falls below %d’s value, market takes
Bearish trend.
Otherwise, if %k’s value rises above %d’s value it indicates Bullish
trend. However, it is very difficult to identify how long will the trend
predicted from FS is valid.

2.1.4.6 Moving Average Convergence Divergence (MACD)

This indicator is calculated based on exponential moving average. To
calculate MACD we first need to compute EMA of 9 intervals and EMA
of 26 intervals. Now starting from 27th interval till 35th interval subtract
EMA (26) from EMA (9) and store the result for further processing.[4]
This value represents upper MACD indicator. With these values now
calculate EMA of previously stored results starting from 36th interval.
This represents lower MACD indicator. Minimum number of intervals
needed for

MACD is 35 and is its only limitation. [5]

Whenever lower MACD value falls below upper MACD value it shows
a trend reversal from Bullish to Bearish. If lower MACD value rises
above its upper MACD value it indicates a trend reversal from Bearish
to Bullish.

13
2.1.4.7 Williams Average (WA)
Williams Average is calculated based on current close, highest
high and lowest low (2.1.2.5). Here is the formula to calculate Williams
Average: In order to calculate this indicator we first subtract current
close from highest high and store the result and call it HC. Now we
subtract lowest low from highest high, store it and call it HL. With
these results we now calculate Williams Average by the following
formula.

WA indicates oversold and overbought conditions. [4] Whenever

it reaches close to 0 it shows oversold condition. If WA value is closer
to 100 it shows overbought condition. [12] We need to identify the
optimal threshold for which this indicator gives accurate results.

2.1.4.8 Fibonacci Series

Fibonacci Series is calculated using the following formula: [4, 5, 16]

Fibonacci series helps us to identify support level or resistance

level of a given stock. Whenever we make a decision based on certain
indicator we can use this indicator to identify we made the right
decision before making a trade.

2.1.4.9 Rate Of Change (ROC)

To calculate ROC, we need to know the current trading price of the
current interval and close price of the interval which is look-ahead
parameter intervals previous to current interval. Here is the formula to
calculate ROC.

14
 Rate of change is directly proportional to the trend of the stock
market. If ROC is lesser than 20 it is an indication that the market is in
bearish trend. If ROC is greater than 80 it is a strong signal to buy
stocks.

2.1.4.10 Artificial Neural Networks (ANN)

This model is a multi-level perceptron model which feeds forward the

result of every computation to the next level of computation. The

author [7] has modified the approach of feed-forward model by adding

pseudo input variables into ANN. First version of ANN propagates

back and forth to provide final results. [7] Author in his approach has

argued that, by modifying the model he could minimize the back and

forth movement of the data to increase optimization of computer’s

computational power. ANN’s are always very less susceptible to noise

points in the data. If a model is trained based on noise points accuracy

of the prediction would decrease drastically due to increase in false

positives. Using this model author is making the decision to hold, sell

or buy stocks at given time. He used a Genetic algorithm to select the

data from the given data set.[20] This selected data will be given as an

15
input to ANN which would classify the labels based on the weight of

computed results. In his results he proves that 2 layer ANN is more

feasible than a single layer perceptron. The author [9] has improved

the model by embedding pattern matching with basic ANN. Pattern

matching algorithms are always highly susceptible to noise.[13]

However, for given patterns the algorithm will detect the re-occurrence

if any. Hence, by embedding this advantage with ANN author has

opened the doors of fortunes. This improved model can now be used

for prediction on data that contains extreme noise (stock market often

is the case). He has also proved the same with his results in the

publication. [9]

2.1.4.11 Hidden Markov Model (HMM)

Hidden Markov Models are statistical models that are used to

determine the most likely sequence of occurrence given a sequence of

data. This model consists of transition matrix, observation matrix and

initial set of observation distribution, denoted as A, B and pi

respectively [21]. This model is used to uncover hidden information

16
behind any set of sequences that has happened. In stock market ticks

of respective intervals are taken into consideration for computing

initial set of observations and observation distribution. Based on the

computed results we then calculate the transition matrix. It is the

transition matrix, which author relied on to calculate the likeliness of

the trading sequence. When he trained this model with ticks whose

results we already know since they are training records. He then

calculated its likeliness. [10] Now he calculates likeliness for a

sequence whose results we don’t know yet. It is based on this

likeliness; we predict the future trend for that particular stock [6]. The

author [10] has used tick prices of current interval as 4 of his

observations. He also considered close price of next interval, relative

to the current interval, as fifth observation for training his model.

Based on these data he computes A, B and pi as discussed previously.

This model now can be used to predict next most likely observation.

[10, 22, 23]

17
 Chapter 3 Methodology

3.1 PROPOSED SYSTEMS

The prediction methods can be roughly divided

into two categories, statistical methods and artificial intelligence
methods. Statistical methods include logistic regression model,
ARCH model, etc. Artificial intelligence methods include multi-
layer perceptron, convolutional neural network, naive Bayes
network, back propagation network, single-layer LSTM, support
vector machine, recurrent neural network, etc. They used Long
short-term memory network (LSTM).

3.1.1 Time Series Analysis

Time series analysis involves the sequential plotting of a set of
observations or data points at regular intervals of time. By studying
previous outcomes and their progression over time, we can make
future predictions with the help of these studied observations.

18
 
The approach to problems related to time series analysis is unique in
its own way. Most machine learning problems utilize a dataset with
certain outcomes to be predicted, such as class labels. This statement
means that most machine learning tasks usually utilize an independent
and dependent variable for the computation of a particular question.
This procedure involves machine learning algorithms analysing a
dataset (say, XX) and using the predictions (YY) to form an assessment
and approach to solving the problem statement. Most supervised
machine learning algorithms perform in this manner. However, time
series analysis is unique because it has only one variable: time. We will
dive deeper into how to solve the stock market price prediction task
with deep learning in the next part of this article. For now, our primary
objective will be understanding the terms and important concepts
required for approaching this task. Let us begin!

The ordered sequence of data points in regular intervals of time

constitutes the primary elements of time series analysis. Time series
analysis has a wide range of applications in the practical world. Hence,
the study of time series is significant to understand, or at the very least,
to gain basic knowledge about what you can expect in the near future.
Apart from the stock market price prediction model that we will build
in the next part of these articles, time series analysis has applications
in economic forecasting, sales forecasting, budgetary analysis, process
and quality control, detection of weather patterns, inventory and utility
studies, and census analysis, among many others.
 
Understanding previous behavioural patterns of data elements is
critical. Consider an example of business or economic forecasting.
When you can extract useful information from the previous patterns,
you can plan the future accordingly. More often than not, the
predictions made with the help of time series analysis and forecasting
yield good results. These will help users to plan and evaluate current
accomplishments.
 
Essential components of time series analysis

19
Time series analysis is all about the collection of previous data to study
the patterns of various trends. By conducting a detailed analysis of
these time series forecasting patterns, we can determine the future
outcomes with the help of our constructive deep learning models.
While there are other methods to determine the realistic results of
future trends, deep learning models are an outstanding approach to
receive some of the best predictions for each concept.
 
Let's now focus on the components of time series forecasting, namely
Trend, Seasonality, Cyclicity, and Irregularity. These four components
of time series analysis refer to types of variations in their graphical
representations. Let us look at each of these concepts individually and
try to gain more intuition behind them with the help of some realistic
examples.

Trend
A Trend in time series forecasting is defined as a long period of time with a
consistent increase or decrease in the data. There may be slight fluctuations in
the data points and elements at various instances of time, but the overall
variation and direction of change remain constant for a longer duration of
time. When a Trend goes from a long duration of constant increase to a long
duration of constant decrease, this is often referred to as a "Changing
Direction" trend.
There a few terminologies used to define the type of trend that we are dealing
with in time series forecasting. A slightly or moderately increasing trend over
a long duration of time can be referred to as an uptrend, while a slightly or
moderately decreasing trend over a long duration of time is called a
downtrend. If the trend is following a consistent pattern of gradually
increasing or gradually decreasing, and there is not much effect in the overall
pattern of the graph, then this trend can be referred to as a horizontal or
stationary trend. The graphical representation shown in the above image has a
downward trajectory over a long period of time. Hence, this image shows the
representation of a downward trend, also called a downtrend.
Let us consider an example to understand the concept of trend better.
Successful companies like Amazon, Apple, Microsoft, Tesla, and similar tech
giants have a reasonably performing upward stock price curve. We have
learned that we can determine this upward rise in stock prices as an uptrend.

20
While successful companies have an uptrend, some companies that are not
performing that well in the stock market have downward trajectories in stock
prices, or a downtrend. Companies that are making a neutral profit rate with
decent amounts of profits and losses at regular intervals are determined as a
horizontal or stationary trend.

 
Seasonality

A pattern that is affected or influenced by seasonal frequency variations are

termed to be seasonal. This component of time series analysis can vary from
time stamps like quarterly, monthly, or half-yearly. However, an important
point to note is that these fluctuations usually take place within the period of a
year. Whenever the frequency is fixed and known and also occurs on a timely
basis, usually within a year, this component of time series analysis is known
as seasonality.
21
To consider a realistic example, think about the sale of certain seasonal fruits.
Fruits like watermelon will have increased sales during the summer season,
while during the winter seasons, the sales of watermelons will gradually
reduce. Similarly, seasonal fruits like apple have higher sales during the
winter seasons in comparison to the other seasons. Ice cream and tender
coconuts are other examples of food products that have increased sales during
the summer season while experiencing a dip in sales during other seasons.
Apart from food products, a period or month like April might have higher
shares of investment and experience a dip in shares until six months later in
October, where it has a peak rise. This pattern can also be regarded as a
seasonality.

Cyclicity
When a pattern exhibits a rise and fall of mixed frequencies, and the graphical
representation has peaks and troughs that occur randomly over a period of
time, it is called a cyclic component. The duration of these occurrences
usually ranges over the period of at least one year. Stock prices of certain
companies that are hard to predict usually have cyclic patterns where they
flourish during a certain period, while having lower profits at other times.
22
Cyclic trends are some of the hardest for our deep learning models to predict.
The graphical representation above shows the cyclic behaviour of house sales
over the span of two decades.
A great realistic example of cyclic behavioural patterns is when a person
decides to invest in their own start-up. During the set-up and progression of
start-ups, every business experiences a cyclic phase. These cyclic fluctuations
often occur in business cycles. Usually, the phases of a start-up would include
the investment stage, which would have a slightly negative impact on our
prices. The next phases could include your marketing and profitable stages,
where you start to earn profits from your successful start-up. Here, you
experience an increase in the graphical curve. However, you will eventually
experience a depreciation phase as well. These will show lesser profits until
you make constant improvements and investments. This procedure begins the
cyclic stage again, lasting for periods of a few years. Rinse and repeat.

Irregularity
Irregularity is a component that is almost impossible to make accurate
predictions for with a deep learning model. Irregularity, or random variations
(as the name suggests), involves an abnormal or atypical pattern where it
becomes hard to deduce the occurrences of the data elements with respect to
23
time. Above is a graphical representation of the Google Stock Price changing
rapidly with an irregular and random pattern; this is hard to read. Despite the
information and data patterns present, the modeling procedure for such kinds
of representations will be hard to crack. The primary objective of the model is
to predict future possibilities based on previous outcomes. Hence, models for
irregular patterns are slightly harder to construct.
To provide a realistic example for irregular patterns, let us analyze the current
status of the world, where many businesses and other industries are affected
on a large scale. The global pandemic is a great example of irregular activity
that is impossible for anyone to predict. These disturbances that occur due to a
natural calamity or phenomenon will affect the trading prices, stock price
charts, companies, and businesses. It is not possible for the model you
construct to detect the occurrences of these situational tragedies. Hence,
irregular patterns are an interesting component of time series forecasting to
analyze and study.

Understanding stationary and non-stationary series

24
Another significant aspect of time series analysis that we will encounter in
this article is the concept of stationarity. When we have a certain number of
data points or elements, and the mean and variance of all these data points
remain constant with time, then we call it a stationary series. However, if
these elements vary with time, we call it a non-stationary series. Most
forecasting tasks, including stock market prices or cryptocurrency graphs, are
non-stationary. Unfortunately, the results obtained by non-stationary patterns
are not efficient. Hence, they need to be converted into a stationary pattern.
The entire data preparation process for stock market price prediction in the
next article is explained with code snippets. Here, we will discuss a few other
topics of importance. Several methods for converting non-stationary series
into stationary series include differencing and transformation. Differencing
involves subtracting two consecutive data points from the higher to lower
order, while transformation involves transforming or diverging the series.
Typically, a log transform is used for this process. For further information on
this topic, refer to the link provided in the image source above.
To test for stationarity in Python, the two main methods that are utilized
include rolling statistics and the ADCF Test. Rolling statistics are more of a
visual technique that involves plotting the moving average or moving
variance to check if it varies with time. The Augmented Dickey-Fuller test
(ADCF) is used to give us various values that can help in identifying
stationarity. These test results comprise some statistics and critical values.
They are used to verify stationarity.

3.1.2 Long short-term memory network:

Recurrent neural networks, also known as RNNs, are a class
of neural networks that allow previous outputs to be used as
25
 inputs while having hidden states. RNNs have issues with the
transfer of long-term data elements due to the possibility of
exploding and vanishing gradients. The fix to these issues is
offered by the Long short-term memory (LSTM) model, which is
also a recurrent neural network (RNN) architecture that is used
to solve many complex deep learning problems. LSTMs are
especially useful for our task of Stock Price Prediction Using
Deep Learning.
The LSTM architecture, with its effective mechanism of
memory cells, is extremely useful for solving complex problems
and making overall efficient predictions with higher accuracy.
LSTMs learn when to forget and when to remember the
information provided to them. The basic anatomy of the LSTM
structure can be viewed in three main steps.
The first stage is the cell state that contains the basic and
initial data to be recollected. The second stage is the hidden state
that consists of mainly three gates: the forget gate, input gate,
and output gate. We will discuss these gates shortly. Finally, we
have a looping stage that reconnects the data elements for
computation at the end of each time step.

Working of LSTM:
LSTM is a special network structure with three
“gate” structures. Three gates are placed in an LSTM unit, called
input gate, forgetting gate and output gate. While information
enters the LSTM’s network, it can be selected by rules. Only the
information conforms to the algorithm will be left, and the
information that does not conform will be forgotten through the
forgetting gate.

The experimental data in this paper are the actual

historical data downloaded from the Internet. Three data sets
were used in the experiments. It is needed to find an optimization
algorithm that requires less resources and has faster convergence
26
 speed.

• Used Long Short-term Memory (LSTM) with embedded layer

and the LSTM neural network with automatic encoder.
• LSTM is used instead of RNN to avoid exploding and
vanishing gradients.

• The historical stock data table contains the information of

opening price, the highest price, lowest price, closing price,
transaction date, volume and so on.
• The accuracy of this LSTM model used in this project is 57%.

Fig. 4: LSTM Architecture

Forget Gate:

A forget gate is responsible for removing information from the cell

state.

• The information that is no longer required for the

LSTM to understand things or the information
that is of less importance is removed via
27
 multiplication of a filter.
• This is required for optimizing the performance of the LSTM
network.
• This gate takes in two inputs; h_t-1 and x_t. h_t-1
is the hidden state from the previous cell or the
output of the previous cell and x_t is the input at
that particular time step.

Input Gate:

1. Regulating what values need to be added to the cell

state by involving a sigmoid function. This is
basically very similar to the forget gate and acts as
a filter for all the information from hi-1 and x_t.
2. Creating a vector containing all possible values
that can be added (as perceived from h_t-1 and
x_t) to the cell state. This is done using the tanh
function, which outputs values from -1 to +1.
3. Multiplying the value of the regulatory filter (the
sigmoid gate) to the created vector (the tanh
function) and then adding this useful information
to the cell state via addition operation.
4.

Output Gate:

The functioning of an output gate can again be broken down to

three steps:
28
 • Creating a vector after applying tanh function to
the cell state, thereby scaling the values to the
range -1 to +1.
• Making a filter using the values of h_t-1 and x_t,
such that it can regulate the values that need to be
output from the vector created above. This filter
again employs a sigmoid function.
• Multiplying the value of this regulatory filter to the
vector created in step 1, and sending it out as a
output and also to the hidden state of the next cell.

The mathematical computations for LSTMs can be interpreted in

many ways. The images shown below are two of the methods in which
the mathematical calculations of LSTMs are carried out. Sometimes
the bias function is ignored, as shown in the second image. However,
we will not cover the intricate details of the first image. You can check
out more on that representation from the link provided in the image
source. The mathematical equations in consideration of the first
LSTM image are pretty accurate for computation purposes.

29
 Where:
X(t)X(t): input vector to the LSTM unit
• F(t)F(t): forget gate's activation vector
• I(t)I(t): input/update gate's activation vector
• O(t)O(t): output gate's activation vector
• H(t)H(t): hidden state vector, also known as output vector of the
LSTM unit
• c˜(t)c̃(t): cell input activation vector
• C(t)C(t): cell state vector
• WW, UU, and BB: weight matrices and bias vector parameters
which need to be learned during training

• # LSTM
• Inputs: dataset
• Outputs: RMSE of the forecasted data
•
• # Split dataset into 75% training and 25% testing data
• size = length(dataset) * 0.75
• train = dataset [0 to size]
• test = dataset [size to length(dataset)]
•
• # Procedure to fit the LSTM model
• # Procedure LSTMAlgorithm (train, test, train_size, epochs)
30
• X = train
• y = test
• model = Sequential ()
• model.add (LSTM (50), stateful=True)
• model. compile (optimizer='adam', loss='mse')
• model.fit (X, y, epochs=epochs, validation_split=0.2)
• return model
•
• # Procedure to make predictions
• # Procedure getPredictonsFromModel (model, X)
• predictions = model.predict(X)
• return predictions
•
• epochs = 100
• neurons = 50
• predictions = empty

31
 32
• # Fit the LSTM model
• model = LSTMAlgorithm (train, epoch, neurons)

Fig. 5: Pre-processing of data

•

• Preprocessing of data

• Overall Architecture
• # Make predictions
• pred = model.predict(train)

3.2 SYSTEM ARCHITECTURE

•
• # Validate the model n = len(dataset)
•
• error = 0
• for i in range(n): error += (abs(real[i] - pred[i])/real[i]) * 100
• accuracy = 100 - error/n
•
Fig. 6: Overall Architecture

33
 Chapter 4 Design

4.1 Structure Chart

A structure chart (SC) in software engineering
and organizational theory is a chart which shows the
breakdown of a system to its lowest manageable levels. They
are used in structured programming to arrange program
modules into a tree. Each module is represented by a box,
which contains the module's name.

34
 Fig. 7: Training and prediction

4.2 UML Diagrams

A UML diagram is a partial graphical

representation (view) of a model of a system under design,
implementation, or already in existence. UML diagram
contains graphical elements (symbols) - UML nodes
connected with edges (also known as paths or flows) - that
represent elements in the UML model of the designed
system. The UML model of the system might also contain
other documentation such as use cases written as templated
texts.

The kind of the diagram is defined by the

primary graphical symbols shown on the diagram. For
example, a diagram where the primary symbols in the
contents area are classes is class diagram. A diagram which
shows use cases and actors is use case diagram. A sequence
diagram shows sequence of message exchanges between
lifelines.

UML specification does not preclude mixing

of different kinds of diagrams, e.g., to combine structural and
behavioral elements to show a state machine nested inside a
use case. Consequently, the boundaries between the various
kinds of diagrams are not strictly enforced. At the same time,
some UML Tools do restrict set of available graphical
elements which could be used when working on specific type
of diagram.

UML specification defines two major kinds of

UML diagram: structure diagrams and behavior diagrams.

35
 Structure diagrams show the static structure
of the system and its parts on different abstraction and
implementation levels and how they are related to each
other. The elements in a structure diagram represent the
meaningful concepts of a system, and may include abstract,
real world and implementation concepts.

Behavior diagrams show the dynamic

behavior of the objects in a system, which can be described as
a series of changes to the system over time.

36
 4.2.1 Use Case Diagram

In the Unified Modelling Language (UML), a use

case diagram can summarize the details of your system's users (also
known as actors) and their interactions with the system. To build
one, you'll use a set of specialized symbols and connectors. An
effective use case diagram can help your team discuss and
represent:

 Scenarios in which your system or application interacts with

people, organizations, or external systems.
 Goals that your system or application helps those
entities (known as actors) achieve.
 The scope of your system.

Fig. 8: Using LMS, LSTM and LSTM with LMS in the system

37
 4.2.2 Sequence Diagram

A sequence diagram is a type of interaction

diagram because it describes how and in what order a group of
objects works together. These diagrams are used by software
developers and business professionals to understand requirements
for a new system or to document an existing process. Sequence
diagrams are sometimes known as event diagrams or event
scenarios.

Sequence diagrams can be useful references for

businesses and other organizations. Try drawing a sequence
diagram to:

 Represent the details of a UML use case.

 Model the logic of a sophisticated procedure, function, or operation.
 See how objects and components interact with each other to complete a
process.
 Plan and understand the detailed functionality of an existing or future
scenario.

38
Fig. 9: Execution based on model selection

39
 4.2.3 Activity Diagram

An activity diagram is a behavioral diagram i.e. it depicts the behavior of a

system.
An activity diagram portrays the control flow from a start point to
a finish point showing the various decision paths that exist while
the activity is being executed.

40
Fig. 10: Execution based on algorithm selection

41
 4.2.4 Collaboration Diagram

Collaboration diagrams are used to show how

objects interact to perform the behavior of a particular use case, or
a part of a use case. Along with sequence diagrams, collaboration
are used by designers to define and clarify the roles of the objects
that perform a particular flow of events of a use case. They are the
primary source of information used to determining class
responsibilities and interfaces.

The collaborations are used when it is essential to

depict the relationship between the object. Both the sequence and
collaboration diagrams represent the same information, but the way
of portraying it quite different. The collaboration diagrams are best
suited for analyzing use cases.

42
Fig. 11: Data transfer between modules

43
 4.2.5 Flow Chart

A flowchart is a type of diagram that represents a

workflow or process. A flowchart can also be defined as a
diagrammatic representation of an algorithm, a step-by-step
approach to solving a task. The flowchart shows the steps as boxes
of various kinds, and their order by connecting the boxes with
arrows.

44
Fig. 12: Flow of execution

45
 4.2.6 Component Diagram

Component diagram is a special kind of

diagram in UML. The purpose is also different from all other
diagrams discussed so far. It does not describe the functionality
of the system but it describes the components used to make those
functionalities.

Component diagrams are used in modeling the

physical aspects of object-oriented systems that are used for
visualizing, specifying, and documenting component-based
systems and also for constructing executable systems through
forward and reverse engineering. Component diagrams are
essentially class diagrams that focus on a system's components
that often used to model the static implementation view of a
system.

Page | 46
 Fig. 13: Components present in the system

Chapter 5 Experiment Analysis

5.1 System configuration

This project can run on commodity
hardware. We ran entire project on an Intel I5 processor
with 8 GB Ram, 2 GB Nvidia Graphic Processor, It also
has 2 cores which runs at 1.7 GHz, 2.1 GHz respectively.
First part of the is training phase which takes 10-15 mins
of time and the second part is testing part which only
takes few seconds to make predictions and calculate
accuracy.
5.1.1 Hardware Requirements:
 RAM: 4 GB
 Storage: 500 GB
 CPU: 2 GHz or faster
 Architecture: 32-bit or 64-bit

5.1.2. Software Requirements:

 Python 3.5 in Google Colab is used for
data pre-processing, model training and
prediction.
 Operating System: windows 7 and above or Linux
based OS or MAC OS

Page | 47
5.1.3 Functional requirements
Functional requirements describe what the software
should do
(the functions). Think about the core operations.

Because the “functions” are established before

development, functional requirements should be written
in the future tense. In developing the software for Stock
Price Prediction, some of the functional requirements
could include:

 The software shall accept the tw_spydata_raw.csv

dataset as input.
 The software should shall do pre-
processing (like verifying for missing data
values) on input for model training.
 The software shall use LSTM ARCHITECTURE as
main component of the software.
 It processes the given input data by producing the
most possible outcomes of a CLOSING STOCK
PRICE.
Notice that each requirement is directly related to what
we expect the software to do. They represent some of the
core functions.

5.1.4 Non-Functional requirements

Product properties
 Usability: It defines the user interface of the
software in terms of simplicity of understanding the
user interface of stock prediction software, for any
kind of stock trader and other stakeholders in stock
market.
 Efficiency: maintaining the possible
highest accuracy in the closing stock prices
in shortest time with available data.

Page | 48
Performance: It is a quality attribute of the
stock prediction software that describes the
responsiveness to various user interactions with
it.

5.2 Sample code

Page | 49
Page | 50
Page | 51
Page | 52
Page | 53
Page | 54
Page | 55
Page | 56
Page | 57
Page | 58
Arima Model

Page | 59
Page | 60
Ad-fuller Test

Page | 61
Page | 62
Page | 63
Predict the closing Stock Price

Page | 64
Machine learning Model
import tensorflow as tf

def Dataset(Data,Date):
Train_Data=Data['Adj. Close'][Data['Date']<Date].to_numpy()
Data_Train=[]
Data_Train_X=[]
Data_Train_Y=[]
for i in range(0,len(Train_Data),5):
try:
Data_Train.append(Train_Data[i:i+5])
except:
pass
if len(Data_Train[-1])<5:
Data_Train.pop(-1)
Data_Train_X=Data_Train[0:-1]
Data_Train_X=np.array(Data_Train_X)
Data_Train_X=Data_Train_X.reshape((-1,5,1))
Data_Train_Y=Data_Train[1:len(Data_Train)]
Data_Train_Y=np.array(Data_Train_Y)
Data_Train_Y=Data_Train_Y.reshape((-1,5,1))

Page | 65
 Test_Data=Data['Adj. Close'][Data['Date']>=Date].to_numpy()
Data_Test=[]
Data_Test_X=[]
Data_Test_Y=[]
for i in range(0,len(Test_Data),5):
try:
Data_Test.append(Test_Data[i:i+5])
except:
pass
if len(Data_Test[-1])<5:
Data_Test.pop(-1)
Data_Test_X=Data_Test[0:-1]
Data_Test_X=np.array(Data_Test_X)
Data_Test_X=Data_Test_X.reshape((-1,5,1))
Data_Test_Y=Data_Test[1:len(Data_Test)]
Data_Test_Y=np.array(Data_Test_Y)
Data_Test_Y=Data_Test_Y.reshape((-1,5,1))
return Data_Train_X,Data_Train_Y,Data_Test_X,Data_Test_Y

Deep Learning Model

def Model():
model=tf.keras.models.Sequential([
tf.keras.layers.LSTM(200,input_shape=(5,1),activation=tf.nn.leaky_r
elu,return_sequences=True),
tf.keras.layers.LSTM(200,activation=tf.nn.leaky_relu),
tf.keras.layers.Dense(200,activation=tf.nn.leaky_relu),
tf.keras.layers.Dense(100,activation=tf.nn.leaky_relu),
tf.keras.layers.Dense(50,activation=tf.nn.leaky_relu),
tf.keras.layers.Dense(5,activation=tf.nn.leaky_relu)
])
return model

Page | 66
model=Model()

tf.keras.utils.plot_model(model,show_shapes=True)

Page | 67
Page | 68
Model Fitting
Compile model

Page | 69
Save model in folder

Page | 70
Webpage Code

from unicodedata import name

import streamlit as st
import yfinance as yf
import pandas as pd
import numpy as np
import cufflinks as cf
from prophet import Prophet
from datetime import datetime
import matplotlib.pyplot as plt
from keras.models import load_model
import pandas_datareader.data as web
from plotly import graph_objects as go
from prophet.plot import plot_plotly, plot_components_plotly
import datetime

# App title

st.set_page_config(page_title="Stock Prediction App")

st.markdown("""<style>#root>div:nth-
child(1)>div>div>div>div>section>div {color:red}
</style>
""",unsafe_allow_html=True)

st.markdown('''
# Stock Price App
Shown are the stock price data for query companies!
**Credits**
- Built in `Python` using `streamlit`,`yfinance`, `cufflinks`, `pandas`
and `datetime`
''')

Page | 71
st.write('---')

# Sidebar
st.sidebar.subheader('Query parameters')
start_date = st.sidebar.date_input("Start date", datetime.date(2005, 1,
1))
end_date = st.sidebar.date_input("End date", datetime.date(2022, 4,
30))
n_years=st.slider("years of predictions: ",1,4)
period=n_years*365

# Retrieving tickers data

ticker_list =
pd.read_csv('https://raw.githubusercontent.com/dataprofessor/s-and-
p-500-companies/master/data/constituents_symbols.txt')
tickerSymbol = st.sidebar.selectbox('Stock ticker', ticker_list) #
Select ticker symbol
tickerData = yf.Ticker(tickerSymbol) # Get ticker data
tickerDf = tickerData.history(period='1d', start=start_date,
end=end_date) #get the historical prices for this ticker

# Ticker information
string_logo = '<img src=%s>' % tickerData.info['logo_url']
st.markdown(string_logo, unsafe_allow_html=True)

string_name = tickerData.info['longName']
st.header('**%s**' % string_name)

string_summary = tickerData.info['longBusinessSummary']
st.info(string_summary)

# Ticker data
st.header('**Ticker data**')

Page | 72
st.write(tickerDf)

@st.cache
def load_data(ticker):
data=web.DataReader(ticker,data_source='yahoo',start=start_date)
data.reset_index(inplace=True)
return data

data=load_data(tickerSymbol)

#describing data
st.subheader('Data from 2005 - Till Now')
st.write(data.describe())

# Time series Data

def plot_row_data():
fig=go.Figure(layout=go.Layout(height=600,width=900))
fig.add_trace(go.Scatter(x=data['Date'],y=data['Open'],name='stock_
open'))
fig.add_trace(go.Scatter(x=data['Date'],y=data['Close'],name='stock_
close'))
fig.layout.update(title_text="Time Series
Data",xaxis_rangeslider_visible=True)
st.plotly_chart(fig)

plot_row_data()

st.subheader('Closing Price vs time chart 100MA ')

ma100=data.Close.rolling(100).mean()
fig=plt.figure(figsize=(12,6))
plt.plot(ma100)
plt.plot(data.Close)
plt.legend()
st.pyplot(fig)

Page | 73
st.subheader('Closing Price vs time chart 100MA & 200MA')
ma100=data.Close.rolling(100).mean()
ma200=data.Close.rolling(200).mean()
fig=plt.figure(figsize=(12,6))
plt.plot(ma100,'r')
plt.plot(ma200,'g')
plt.plot(data.Close,'b')
plt.legend()
st.pyplot(fig)

# Bollinger bands
st.header('**Bollinger Bands**')
qf=cf.QuantFig(tickerDf,title='First Quant
Figure',legend='top',name='GS')
qf.add_bollinger_bands(periods=20,boll_std=2,colors=['cyan','gray'],
fill=True)
# qf.add_volume(name='volume',up_color='green',down_color='red')
fig = qf.iplot(asFigure=True)

fig.layout.update(title_text="Time Series
Data",xaxis_rangeslider_visible=True)
st.plotly_chart(fig)

#forecasting
df_train=data[['Date','Close']]
df_train=df_train.rename(columns={"Date":"ds","Close":"y"})

m=Prophet()
m.fit(df_train)

future=m.make_future_dataframe(periods=period)
forecast=m.predict(future)

st.subheader("Forecast data")

Page | 74
st.write(forecast.tail())

st.write("Forecast data")
fig1=plot_plotly(m, forecast)
st.plotly_chart(fig1)

st.write("Forecast components")
fig2=plot_components_plotly(m,forecast)
st.plotly_chart(fig2)

data_training=pd.DataFrame(data['Close'][0:int(len(data)*0.70)])
data_testing=pd.DataFrame(data['Close']
[int(len(data)*0.70):int(len(data))])

from sklearn.preprocessing import MinMaxScaler

scaler=MinMaxScaler(feature_range=(0,1))

data_training_array=scaler.fit_transform(data_training)

#load model
model =load_model('keras_model.h5')

#testing part

past_100_days=data_training.tail(100)
final_df=past_100_days.append(data_testing,ignore_index=True)
input_data=scaler.fit_transform(final_df)

x_test=[]
y_test=[]

Page | 75
for i in range(100,input_data.shape[0]):
x_test.append(input_data[i-100:i])
y_test.append(input_data[i,0])
x_test,y_test=np.array(x_test),np.array(y_test)

# making predictions
y_predicted=model.predict(x_test)
scaler=scaler.scale_
scaler_factor=1/scaler[0]
y_predicted=y_predicted*scaler_factor
y_test=y_test*scaler_factor

#final graph

st.subheader('Predictions vs Original ')

fig2=plt.figure(figsize=(12,6))
plt.plot(y_test,'b',label='Original Price')
plt.plot(y_predicted,'r',label='Predicted Price')
plt.xlabel('Time')
plt.ylabel('Price')
plt.legend()
st.pyplot(fig2)

Web Screen

 Main Screen

Page | 76
 Company Information

Page | 77
Page | 78
Page | 79
 Forecast Component

Page | 80
 Forecast Component

Page | 81
Chapter 6 Conclusion And
Future Work

6.1 Conclusion

In this project, we are predicting

closing stock price of any given organization, we
developed a web application for predicting close
stock price using LMS and LSTM algorithms for
prediction. We have applied datasets belonging to
Google, Nifty50, TCS, Infosys and Reliance Stocks
and achieved above 95% accuracy for these datasets.

6.2 Future work

Page | 82
 We want to extend this application for predicting
cryptocurrency trading.
 We want to add sentiment analysis for better analysis.

Page | 83