STOCK MARKET PREDICTION USING PYTHON

Submitted in partial fulfillment of the requirements for the award

of Bachelor of science in computer science

BY
MS.B.JOTHILAKSHMI(39290044)
MS.M.SHALINI (39290093)

DEPARTMENT OF COMPUTER
SCIENCE SCHOOL OF COMPUTER
SCIENCE

SATHYABAMA
2022 INSTITUTE OF SCIENCE AND TECHNOLOGY

(DEEMED TO BE UNIVERSITY)
Accredited with Grade "A" by JEPPIAR NAGAR, RAJIV GANDHI
SALAI, CHENNAI-600 119
May-2022
 SATHYABAMA
INSTITUTE OF SCIENCE AND TECHNOLOGY

(DEEMED TO BE UNIVERSITY)

Accredited with Grade “A” by NAAC

JEPPIAAR NAGAR , RAJIV GANDHI SALAI CHENNAI -600119

www.sathyabama.ac.in

DEPARTMENT OF COMPUTER SCIENCE

BONAFIDE CERTIFICATE

This is to certify that the Project Report is the bonafide work of

B.JOTHILAKSHMI (39290093) and M.SHALINI(39290093) who carried out
the project entitled "STOCK MARKET PREDICTION USING PYTHON " under
our supervision from November 2021 to May 2022.

Internal Guide

(Mrs. M.SELVI, ME.,PH.D.,)

Head of the department

(Dr. L. LAKSHNMANAN M.E., Ph.D.,)

Submitted for viva voce examination held on

Internal Examiner External Examiner

 DECLARATION

I , M.SHALINI here by declare that the Project Report entitled “STOCK

MARKET PREDICTION USING PYTHON” done by me under the guidance of
Dr. M.SELVI is submitted in partial fulfillment of the requirements for the
award of Bachelor of Engineering / Technology degree in Computer
science.

DATE:

PLACE: CHENNAI SIGNATURE OF THE CANDIDATE

 ACKNOWLEDGEMENT

I am pleased to acknowledge my sincere thanks to Board of Management

Of SATHYABAMA for their kind encouragement in doing this project
and for Completing it successfully. I am grateful to them.

I convey my thanks to Dr. T. SASIKALA, M.E., Ph.D., Dean, School of

Computer and Dr. S . VIGNESHWARI M.E., Ph.D., and Dr. L.
LAKSHNMANAN M.E.,
Ph.D., Head of the Department for providing me necessary support and
details at the right time during the progressive reviews.

I would like to express my sincere and deep sense of gratitude to my

Project Guide Mrs. M.SELVI, ME., PH.D., for his valuable guidance,
suggestions andconstant encouragement paved way for the successful
completion of my project work.

I wish to express my thanks to all Teaching and Non-teaching staff

members of the
Department of computer science who were helpful in many ways for the

Completion of the project.

 ABSTRACT

The prediction of a stock market direction y serve as an early

recommendation system for short -term investors and early financial distress
warning system for a long term shareholder. Forecasting accuracy is the
most important factor in selecting any forecasting methods. Research efforts
in improving the accuracy of forecasting models are increasing since the last
decade. The appropriate stock selections those are suitable for investment is
a very difficult task. The key factor for each investor is to earn maximum
profits on their investments. In this paper support vector machine algorithm
(SVM) is used. SVM algorithm is very specific type of learning algorithms
characterized by the capacity control of the decision function, the user of the
kernel functions and the scarcity of the solution. In this paper, we investigate
the predictability of the financial movement with SVM. To evaluate the
forecasting ability of SVM, we compare its performed with decision trees.
These methods are applied on 2 years of data retrieved from Yahoo finance.
The result will be used to analyze the stock prices and their prediction in
depth in future research efforts.
 TABLE OF CONTENT

CHAPTER TITLE PAGENO

ABSTRACR
LIST OF FIGURES
LIST OF ABBREVIATION

1 INTRODUCTION

2 2.1 Literature survey

2.1.1 Stock price prediction
2.1.2 Neural network
2.1.3 Long short-term memory (LSTM)
2.1.4 Evolution algorithm
2.2 System analysis
2.2.1 Existing system
2.2.2 Proposed system

3 3.1 Aim of the project

3.2 Scope and Objectives of the project
3.2.1 Research
3.2.2 Application
3.3 Overview of the project

ⅵ
4 Technologies used
4.1 Hardware required
4.2 software required

5 Methodology design and implementation

5.1 System Architecture
5.2 Problem framing
5.3 Robust design
5.4 Performance evaluation
5.4.1 Motivation
5.4.2 Definitions
5.5 Model
5.6 Application design
5.6.1 User group
5.6.2 User journey
5.7 Research implementation
5.7.1 Stock Price Data Collection
5.7.2 Data Pre-processing
5.7.3 Model
5.7.4 Training
5.7.5 Saving Trained Model
5.7.6 Predicting Stock Price
5.7.7 Save predictions
5.8 Application implementation
ⅶ
 5.8.1 Stock Information Collection
5.8.2 Google charts
5.8.3 FB login via firebase authentication
5.8.4 Components
5.9 Findings
5.9.1 General findings
5.9.2 Prediction approach findings

6 Testing
7.1 Unit testing
7.2 Tool used for testing

7 Result and Discussion

8.1 Accuracy of stock price prediction
8.2 democratization of machine learning

8 Conclusion and future work

References

Appendix
A. SOURCE CODE
B. SCREENSHOTS

ⅷ
 LIST OF FIGURES

FIG NO FIGURES NAMES PAGE NO

3.3 LINEAR REGRESSION METHOD

5.1 SYSTEM ARCHITECTURE

5.2 WORK FLOW

5.3 ARCHITECTURE OF RECURRENT NEURAL

5.4 USER GROUP NETWORK

5.5 USER JOURNEY

5.8 COMPONENT DIAGRAM

ⅸ
 LIST OF ABBREVIATION

S.NO ABBERVIATION EXPANSION

1 LSTM Long short-term

memory

2 GRU Gated recurrent unit

ⅹ
 CHAPTER 1
INTRODUCTION

Forecasting of stock market is a way to predict future prices of stocks. It is a

long time attractive topic for researcher and investors from its existence. The
Stock prices are dynamic day by day, so it is hard to decide what is the best
time to buy and sell stocks. Machine Learning provides a wide range of
algorithms, which has been reported to be quite effective in predicting the
future stock prices. In this project, we explored different data mining
algorithms to forecast stock market prices for NSE stock market. Our goal is
to compare various algorithms and evaluate models by comparing prediction
accuracy. We examined a few models including Linear regression, Arima,
LSTM, Random Forest and Support Vector Regression. Based on the accuracy
calculated using RMSE of all the models, we predicted prices of different
industries. For forecasting, we used historical data of NSE stock market and
applied a few preprocessing methods to make prediction more accurate and
relevant. The ultimate goal of our application is to serve retail investors as a
third party investment tool that uses machine learning to help them navigate
in the fast- changing stock market. The project aims to introduce and
democratize the latest machine learning technologies for retail investors. No
prediction is 100% accurate. Therefore, the upper bound and lower bound of
the stock prices will be displayed to illustrate the trading range the investors
should be looking this application serves. Stock market prediction ad analysis
are some of the most difficult jobs to complete. There are numerous causes
for this, including market volatility and variety of other dependent and
independent variables that influence the value of a certain stock in the
market. These variable make it extremely difficult for any stock market expert
to anticipate the rise and fall on the market with great precision. Machine
learning is a strong algorithm the most recent market research and stock
market prediction advancements have begun to include such approaches in
1
analyzing stock market data. Implementation of analyzing and forecasting the
stock price in python using various machine learning algorithms. Stock price
prediction using machine learning helps to discover the future value of a
company stock and other financial assets traded on an exchange. The entire
idea of stock prices is an gain significant profits. Stock

2
market helps companies to raise capital . It helps to generate personal
wealth. Stock market server as an indicator of the state of the economy. It is
a widely used source for the people to invest money in companies with high
growth.

3
 CHAPTER 2
2.1 LITERATURE SURVEY

2.1.1 STOCK PRICE PREDICITION

From the research paper “Machine Learning in Stock Price Trend
Forecasting” written by Y. Daiand Y. Zhang In Stanford University, they used
features like PE ratio, PX volume, PXEBITDA, 10-day volatility, 50-day moving
average, etc. To predict the next-day stock price and long-term stock price.
The machine learning algorithms used in the research are Logistic
Regression, Gaussian Discriminant Analysis ,Quadratic Discriminant Analysis
,and SVM. The accuracy ratio is defined as the number of days that the
model correctly classified the testing data over the total number of testing
days. With the short-term model predicting the next day stock price, it has
very low accuracy, the Quadratic Discriminant Analysis is the best among all
models, it scored a 58.2% accuracy. With the long -term model predicting
the next n days stock prices, the longer the time frame, the better in the
accuracy for SVM. With a time window of 44 days, the VM model
’saccuracyreached79.3%. Apart from that, It was found that by increasing
the number of features, the accuracy increased. When all of the 16 features
were used, the accuracy of the model reached 79%, while it fell to 64% when
only 8 features were used, and 55% if only 1 feature was used. Our project
will also investigate how the time frame would affect the accuracy of price
predictions of different models. As models have to reach a certain
significance for the users to work as a reference, it is essential for us to
optimize our model to figure out what the optimal parameters and model
structure are for our stock price prediction purpose. B.Wanjawa and L.
Muchemi demonstrated the potential in predicting stock prices using ANN, as
shown in the research paper “ANN Model to Predict Stock Prices at Stock
Exchange Markets”[4]. They used 70% of the training data to predict the
stock prices for the next 60 days. Through optimizations, they were able to

4
predict the actual closing prices within 0.71% mean absolute percentage
error (MAPE), with the highest variance-3.2% among all of the 62 days. This
demonstrated a high potential for using machine learning to accurately
predict stock prices. This is one of the key components in our application
where algorithms have to be designed to have high accuracy, such that the
platform could be useful for retail investors.

5
2.1.2 NEURAL NETWORK
A neural network attempts to learn a function that maps the input
features to the output predictions, serving as a universal function
approximator. It consists of a network of neurons, each of which represents a
weighted sum of inputs. Outputs from neurons are fit into activation
functions which introduce non-linear to the system, and the n passed to
some other neurons. In a typical dense feedforward neural network, the
network consists of layers of neurons stacked together, with neurons
between individual layers fully connected.

2.1.3 LONG SHORT-TERM MEMORY[LSTM]:

Long short-term memory [8] was first introduced by Hochreiter and
Schmid huber in 1997 to address the a for mentioned problems. Long-short
term memory tackles the problem of learning to remember information over
a time interval, by introducing memory cells and gate units in the neural
network architecture. A typical formulation involves the use of memory
cells, each of which has a cell state that store previously encountered
information. Every time an input is passed into the memory cell, and the
output is determined by a combination of the cell state (which is a
representation of the previous information), and the cell state is updated.
When another input is passed into the memory cell, the updated cell state
and the new input can be used to compute the new output.

2.1.4 EVOLUTION ALGORITHM:

Researches have shown that large-scale evolution can auto-
generate neural network model architectures and hyperparameters with
performance comparable with state-of-the-art human-designed models. In a
research in 2017 , a large-scale evolution for discovering image classification
neural networks was run. It started with a huge population of randomized
simple1-layermodels, then slowly evolved the population by removing a poor
model and generating a new model by mutating some parameters of a good
6
model in each iteration. After hundreds of hours of running the algorithm
with huge computing power, most models in the population achieved state-
of-the-art results on CIFAR datasets. In each iteration, only a simple mutation
that changed 1 parameter was

7
applied, which allowed searching in a large search space. The paper showed
the possibility of finding good models by using lots of computational power to
replace human- machine learning experts and has set the foundation of
learning with AutoML.

2.2 SYSTEM ANALYSIS

2.2.1 EXISTING ANALYSIS :

In the finance world stock trading is one of the most important

activities. Stock market prediction is an act of trying to determine the future
value of a stock other financial instrument traded on a financial exchange.
This paper explains the prediction of a stock using Machine Learning. The
technical and fundamental or the time series analysis is used by the most of
the stockbrokers while making the stock predictions. The programming
language is used to predict the stock market using machine learning is
Python. we propose a Machine Learning (ML) approach that will be trained
from the available stocks data and gain intelligence and then uses the
acquired knowledge for an accurate prediction. In this context this study
uses a machine learning technique called Support Vector Machine (SVM) to
predict stock prices for the large and small capitalizations and in the three
different markets, employing prices with both daily and up- to-the-minute
frequencies.

2.2.2 PROPOSED SYSTEM :

In this project, we have made a time-series analysis and it doesn’t
need n- fold cross validation. Methodology since it’s sequential data. We split
our dataset in train and test data. Top 80 percent of data will be Train data
and the remaining will be test. The prediction model, which is based on SVM
and independent analysis, combined called SVM-ICA, is proposed for stock
market prediction. various time series analysis models are based on machine
learning. The SVM designed to solve regression problems in non-linear
8
classification and time series. In this proposed system, we focus on predicting
the stock values using machine learning algorithms like Random Forest and
Support Vector Machines.

9
 CHAPTER 3

3.1 AIM OF THE PROJECT:

Stock market prediction means determining the future
scope of the market . The prediction of a stock market direction y serve as
an early recommendation system for short -term investors and early financial
distress warning system for a long term shareholder. Forecasting accuracy is
the most important factor in selecting any forecasting methods. Research
efforts in improving the accuracy of forecasting models are increasing since
the last decade. The appropriate stock selections those are suitable for
investment is a very difficult task. The key factor for each investor is to earn
maximum profits on their investments .In this paper support vector machine
algorithm (SVM) is used. SVM algorithm is very specific type of learning
algorithms characterized by the capacity control of the decision function ,
the user of the kernel functions and the scarcity of the solution .

3.2 SCOPE AND OBJECTIVES OF THE PROJECT:

The project aims to introduce and democratize the latest
machine learning technologies for retail investors. No prediction is 100%
accurate. Therefore, the upper bound and lower bound of the stock prices
will be displayed to illustrate the trading range the investors should be
looking at. This application serves as a supplementary quantitative tool for
investors to see the market at a different perspective with the help of
technology. This project is divided into 2 parts, namely are search
component and an application component, aiming to provide retail investors
with stock price predictions using different machine learning models in a
good user experience way for reference.

3.2.1 RESEARCH:

This project will investigate how different machine learning

techniques can be used and will affect the accuracy of stock price
10
predictions. Different models, from linear regression to dense and recurrent
neural networks are tested. Different hyperparameters are also tuned for
better performance These arch space for all neural network architectures
and hyperparameter combinations is huge, and with limited time in

11
 conducting this project, apart from manually trying combinations, the team
optimizes the models with evolution algorithm, replicating AutoML
techniques from other researches with promising results in the financial
context.

3.2.2 APPLICATION:
This project aims to provide stock price predictions based on the latest
machine learning technologies to all retail investors. A mobile web application
is developed to provide predictions in an intuitive way. Different models’
performance and accuracy can also be compared. The application also serves
as another user interface (UI) in visualizing results from the research apart
from Jupyter notebooks with lots of tables and graphs

3.3 OVERVIEW OF THE PROJECT:

There are over 2.2 million Hong Kong stock investors, who contributed
about 15% of the cash market trading value in 2016. The total cash market
trading turn over is around HK$1.6trillion. In particular, retail investors have
made buy or sell investment decisions worth a total turn over of $240 billion
for the year of 2016 [1].In Hong Kong, there are a lot of investment decisions
that involve a large sum amount of money being made. Retail investors spend
a lot of time finding investment opportunities. Wealthier investors could seek
professional financial advisory services, but for typical retail investors, the
costs are prohibitive. Thus, retail investors have to figure out the market
themselves and make informed decisions on their own. This makes
investment very stressful in modern societies.
Unfortunately, humans are irrational in nature. Without quantitative, data-
driven models, decisions get swayed by cognitive biases or personal
emotions, resulting in unnecessary losses. Even if investors are cautious
enough, most do not have sufficient skills to process a huge volume of data

12
required to make good judgments. Institutional investors rely on
sophisticated models supported by technologies to avoid traps, but retail
investors do not have access to such technologies and often find themselves
falling behind the market. Without access to quantitative and data-driven
models, one obvious approach retail investors could use to evaluate the
market is through simple indicators, for example, linear regression and
exponential moving average (EMA) (Figure 1.1).Two important

13
indicators are 20-day EMA and 50-day EMA. When the 20-day EMA rises
above the 50- day EMA, the stock is likely to trend upward, and vice versa.
Another obvious approach retail investors might use to predict the stock
market is to draw a linear regression line that connects the maximum or
minimum of candle sticks.

Figure 3.3 Linear regression method to evaluate and predict the market trend

14
Inspired by the increasing popularity of deep learning algorithms for
forecasting application, these algorithms might serve as potential tools to
find hidden patterns in the trend of stock prices, this information could be
useful to provide extra insights for retail investors when making investment
decisions. Therefore, this final year project aims to investigate the usefulness
of deep learning algorithms in predicting stock prices and democratize such
technologies through an easy to use interface for the general public.

15
 CHAPTER 4

TECHNOLOGY USED:

 MACHINE LEARNING MODELS USED: FB Prophet

 Python for backend
 Flask framework for integration of frontend and backend
 JavaScript frame work with CSS and HTML for front end
 Dygraphs for plotting

4.1 HARDWARE REQUIRED:

• Hard disk : 500 GB and above.
• Processor : i3 and above.
• Ram : 4GB and above.
•
4.2 SOFTWARE REQUIRED:
 Operating System : Windows 10 (64 bit)
• Programming Language : Python
• Tool : Jupyter

16
 CHAPTER 5
METHODLOGY DESIGN

5.1 SYSTEM ARCHITECTURE:

The architecture of the system follows a client-server model ,where the

server and the client are loosely coupled.

Figure 5.1 System Architecture Diagram

17
After relevant stock data are retrieved from the third-party data provider
through the cloud, the backend pre-processes the data and builds the
models. After that, predictions are made and the prediction results will be
stored on another cloud, which can be retrieved from the mobile application.
The advantages of the loosely coupled architecture include improved
scalability and ease of collaboration. The workload for the cloud which serves
the models and the one which serves the mobile application will be very
different. One cloud serves the model prediction results, which are simple
text files; another cloud serves the mobile application with a lot of rich user
content such as images and large UI libraries. Having two clouds to adapt to
two different demand patterns is more efficient, especially since cloud
providers these days usually serve content on demand. Also, these parathion
allows different team members in the team to focus on different parts after
agreeing on a common interface. It speeds up development as team
members responsible for different parts of the system do not need to take
care of the under lying implementation details. Also, it is easier to swap out
different components, e.g. to replace the models the team could simply
18
make changes to the backend, while the frontend remains unaffected.

19
5.2 PROBLEM FRAMING:
The problem of the project is set to predict the stock price for
the next 10 business days. “10days” is chosen as the time frame as short
term price movements tend to depend more on trend momentum and price
pattern, while long term price movements depend on the fundament also a
stock (e.g. company management capabilities, revenue model, market
demand, macroeconomic factors, etc.). The loss function of the training
algorithm is the mean squared error of the 10 predicted stock prices. The
training algorithm or optimizer is set to minimize its value, and it serves as
the basic performance metric for comparing different models. Other scores
are defined to provide more in-depth in sights on a model predictability
performance and finance-domain-based comparisons between models for
investors. Two different prediction approaches are mainly tested, predicting
the stock prices for the next10 days directly and predicting the stock price of
the next day 1 at a time. It is suspected that the two different problem
framing approaches will result in different abstractions learn performance
for different use-cases.

5.3 ROBUST DESGIN:

For the research side, the system is designed to be as robust as
possible to facilitate model testing. Each model can be defined by a pair of
model options and input options, specifying the model configurations and
the input sit takes. This accelerates the process of testing out different
model and/or input configuration combinations.

5.4 PERFORMANCE EVALUATION:

Model configurations and the inputs it takes. This accelerates
the process of testing out different model and/or input configuration
combinations.

20
 5.4.1 MOTIVATION:
Apart from the mean squared error that a model tries to minimize,
different finance-specifics cores are introduced to evaluate and compare
performance of different models, namely model accuracy score, model trend
score and stock buy/sell score. The
scores are also designed to convey useful and meaningful messages to help investors

21
understand a stock and make investment decisions.

5.4.2 DEFINITION:
In this project, the test set is defined as the last 100 days stock price.
To clearly explain the performance evaluation rationale, the following
symbols are defined.

“Snakes” is defined as 10-day disjoint prediction segments in the test set,

which will be a set of 10“snake”. It includes the actual prices and the
predicted prices for the last 100days.

Specifically, Snakes are defined below:

It is named as “Snakes” because intuitively the 10-day disjoint segments

look like snakes when being plotted on a graph of historical prices.

5.5 MODEL:

Different common neural network models are tested, including dense neural
network, simple recurrent neural networks (RNNs), Long short-term memory
networks(LSTMs)and gated recurrent unit networks (GRUs).Different model
architectures are tested by changing the number of hidden layers, the
number of hidden units per hidden layer, and the activation function or
recurrent activation function used in each hidden layer. All recurrent neural
networks, RNNs, LSTMs, and GRUs, are set to have the recurrent neural
networks same high-level architecture (Figure2.2), a stack of recurrent layers
by passing

22
the full output sequence to the next layer, followed by a stack of dense layers.

Figure 5.5 Example of the common high-level architecture

5.6 APLICATION DESIGN

5.6.1 USER GROUP:
Users are separated into two groups, normal users and advanced
users. For users that would like to know about the historical (test set)
performance of a model and more information behind the machine learning
models like the architecture and inputs, they can enable advanced user
mode in the settings page to view those details in each individual stock

23
page.

24
5.6 Functionality accessible by normal users and advanced user

25
5.6.2 USER JOURNEY:

Figure 5.6 User Journey

First of all, users need to login to use the system, as there will be
customization options for different users. Since users might not want to
create a separate account just for our application, it will more convenient if
users can log in with their existing social media accounts. In particular,
Facebook login is a good option, since there are over 2 billion users
worldwide. Thus, it might be possible to reach a larger market by adopting
Facebook login. Only the very basic user information like the user name will
be collected by the system. For normal users (advanced user mode
disabled), after logging into the system, they can view the stock list and
search from it by stock name.

5.7 RESARCH IMPLEMENTATION:

All machine learning-related code are written in Python. Neural
networks are implemented with Keras while linear regression model is
implemented with scikit-learn
.
5.7.1 STOCK PRICE PREDICITION:
Data is collected from Alpha Vantage Stock Price API [18].It offers up to
20 years of daily stock price information on S&P500 stocks. A Python script is
written to retrieve stock prices of different stocks automatically. The retrieved
stock prices are stored as .csv files in a local folder during development and
testing. In deployment, the downloaded stock price data will be transformed
26
into a 2D JavaScript array and uploaded to Firebase Cloud Storage
immediately. A cron job that launches the data-fetching and data-uploading
script is scheduled to run every 8p.m. (EDT) after NYSE and NASDAQ are
closed.

27
5.7.2 DATA PRE-PROCESSING:
3 Python scripts are written to transform the raw stock prices
(.csv files) into feature vectors, for training, predicting and testing
respectively. The scripts take the input options and the raw stock prices as
inputs and produce the correct features by building the lookback arrays and
the moving averages. It concatenates the features into the final feature
vectors, which will be passed to the model for training or testing. The 3
scripts share common operations in building a dataset except the output size
and the range of dates to build from, so common functions are written to
centralize the logic instead of repeating the same index-calculation-intensive
work across functions . Is frequently used for machine learning tasks
because it is much for performant than Python lists ,as NumPy arrays are
implemented as densely packed lists, instead of a dynamic array where the
elements are not stored contiguously. Pandas is a popular framework for pre-
processing time series data. It has various utilities for reading raw input files
such as .csv and transforming time series data to the correct format. Pandas
uses NumPy as the underlying data structure, so it is very convenient to inter
operate between the two.

5.7.3 MODEL:
A model base class is used as a common interface for all machine
learning models. All models then have their own model class, specifying
model-specific details like methods to build the model, train the model, use
the model and save the model.
To decouple model configurations from software code to provide flexibility
and robustness and save engineering effort as mentioned , each model is
defined by a JSON object, which specifies the model’s architecture and
hyperparameters with model options and the model inputs with input options.
A corresponding model can then be created by passing the object to the
model class constructor. The model options specify which machine learning
model to use, and the hyperparameters for the model like the number of
28
hidden layers, the number of hidden units, activation functions used, as well
as optimization algorithms and loss functions. Some example model options
are in Appendix A.

29
5.7.4 TRAINING:
In training, a randomized initial model is first generated from the
model options definition. A training set is generated by the build training
dataset script, which generates the training set features from the input
options and the raw stock price data. Then, the data is fed into the model for
training.

5.7.5 SAVING TRAINED MODEL:

All trained models are saved for predicting stock prices in the
future. Keras models are saved inh5 format, and scikit-learn models are
saved with a Python library named pickle. A dedicated saving format is
designed (Appendix C), such that same models (same hash for same model
options and input options) for different stocks are saved in the same directory
with no collision.

5.7.6 PREDICTING STOCK PRICE:

When predicting stock price ,the saved model will first be
loaded. Then, a feature vector specified by the input options is built with the
build predict dataset script, which is the same as the build training dataset
except it returns a flatten 1D feature vector. The feature vector is inputted
into the model to predict stock price. For 10-day predict, the predictions are
directly out putted. For 1-day predict, the predicted stock price is appended
to the raw dataset as if it happened before, then a new feature vector is
generated for predicting the stock price for the day after, the process is
repeated to predict the stock prices for all next 10 days.

5.7.7 SAVE PREDICITIONS:

For each stock, a prediction file can be generated from the save
predictions script. It includes all the data and results that the application
needs to display, including all 10-day predictions from all models, both 1-day
30
predict test set and snakes test, and the model options and input options for
each model. The saved predictions file is then saved to Firebase Cloud
Storage and served to the application. During development, the saved
predictions file is saved in a local directon.

31
 5.8 APPLICATION IMPLEMENTATION

5.8.1 STOCK INFORMATION COLLECTION:

Company information is collected from the IEX Stock API
[25]. A Firebase Cloud Function [26] is written to get the data from the API
and store it in Firebase Cloud Fire store [27], which the application will
access through another Firebase Callable Cloud Function.

5.8.2 GOOGLE CHARTS:

Google Charts [34] is used to plot the stock prices and the
predictions. It provides a simple and separate set of APIs that does not
depend on other libraries while maintaining customizability. Also, it
integrates nicely with the user interface which follows the Material Design,
which improves the overall user experience.

5.8.3 FB LOGIN VIA FIREBASE AUTHENTICATION TECHNOLOGY:

To facilitate user logins ,the popular Facebook login service
powered by Firebase Authentication is used, which provides a set of rich set
of APIs to interact with a range of authentication providers, including
Facebook Login, and integrates well with other Fire base services that the
system relies on. It is also convenient to add other authentication service
providers in the future.

5.8.4 COMPONENT:
The whole application is broken down into the following hierarchy of
components. At the top level, the App component brings everything
together. If the user is not logged in he/she will be directed to the login page
(Login Page).The router component (Router) controls where the user will end
up at, including the home page(Home Page), which is the default starting
point for users with a list of stocks and a search bar, all broken down into
separate components Other components include the details page (Detail
32
Page), which is where details about a stock price, including the stock price
chart, a list of models with a model score attached to each model, along with
a buy/sell score that indicates the overall predictions for whether the stock
should be purchased or sold, are included in.

33
Figure 5.8 Component Diagram

34
5.9 FINDINGS
5.9.1 GENERAL FINDINGS:
The following are some general findings from testing out different
machine learning models.

Figure 5.9 1-day interval historical predictions (GE, Dense Neural Network)

35
From Figure 5.9, it shows that the 1-day interval historical predictions line

follows closely with the historical prices. The graph looks like the prediction

line is just 1 day shifting from the historical prices, similar to a shifted and

smoothed out historical prices line. Therefore, the shape of the historical

predictions line is similar to the shape of the exponential moving averages

(EMA), where the price changes from ttot+1 heavily depends on the direction

and Magnitude of changes fromt-1tot, followed by decreasing importance

from earlier historical prices. Other models in predicting stock prices of other

stocks also show similar results. Although price reflects all available

information, the magnitude of price changes in the future might need other

data for forecasting purpose, such as market sentiment company

announcement, retail and institutional investors’ attention on the company,

etc. This is one of the possible explanation of why the 10-day interval

prediction might have a large difference to actual values as there are

potential shifts in market momentum. Therefore, the price might be too

compact and other information is required to make a more accurate

prediction.

36
Figure 5.9 10-day interval historical predictions (GE, Dense Neural Network)

37
From Figure 5.9 it shows that the 10-day interval historical predictions line do
not follow closely with the historical prices but could demonstrate the trend.
For example, historical predictions 1, 2, 3, 4, 7, 8, 9, 10 provided insights on
the correct market direction, yet the magnitude did not match the actual
price movements. A possible reason for this error can be the 10-day interval
prediction has to predict more values while having fewer data compared to
the case of 1-day interval prediction, which for 1-day interval prediction, data
of close prices until previous day are available. Therefore, a longer period of
interval prediction could subject to greater changes in market fundamentals,
including market news, macroeconomic factors, earning reports, etc. Other
models in predicting stock prices of other stocks also show similar results.

5.9.2 PREDICTION FINDINGS

As mentioned in approaches are tested in predicting the stock prices

for the next 10days, predicting all 10-day stock prices directly and predicting

each stock price one at a time. The 2 different approaches frame the

problem totally differently, which introduces a significant language bias.

According to the results (e.g. Figure 6.2a and 6.2b), for most stocks, most

models that predict10-day stock prices directly have a higher error than

predicting individual stock price. However, the errors in predicting different

days in the future are relatively constant for models thatpredict10-day stock

prices directly, while the error increases with the time from now for models

that predict stock prices one day at a time.

38
Figure 5.9 Prediction error in predicting stock price at different future
dates(GOOGL, 10-day predict)

39
 CHAPTER 6
TESTING

6.1 UNIT TESTING:

The unit test module from Python is used to implement all unit
tests, as it is available by default in Python and integrates well with existing
Python codes. Unit tests are done for the build dataset script, which
transforms the raw input data into feature vectors usable for training and
testing, as well as models core calculations. Unit tests are conducted
because the components are error-pron, calculation intensive. Also, they
exhibit garbage-in-garbage-out properties, that the model will be completely
wrong if it receives the wrong input, and if the model scores are wrong, the
final buy-sell recommendation will be totally incorrect.. Combinations of
input options are tested, including n-day stock price look back as well as n-
day moving average. Correctness is ensured by asserting the feature
vectors’ shapes, as well as starting and ending elements.

6.2 TOOL USED FOR TESTING:

Various tools have been used to assist in the development of
the mobile application. In particular , Chrome Mobile Emulator is used to
simulate the mobile view while developing the mobile application on
desktop/laptop computers. After the application is deployed to the cloud,
mobile phones with different operating systems and browsers, including
Google Pixel running Android 9 (Google Chrome) and iPhone 7 running iOS
12.1 (Safari), are used to verify the user experience is consistent across
different devices with different resolutions .

40
 CHAPTER 7

RESULT AND DISCUSSION

RESULT:

FIG 7.1 HOME PAGE-LOGIN PAGE

This is the login page of the project . Enter the http link in the google
chrome ,then the login page will be appear .Enter the company tricker
symbol and submit by clicking submit button

41
 FIG 7.2 GRAPH

After submitting the company tricker symbol, it display the prediction graph
of the stock market.

42
 DISCUSSION:

7.1 ACCURACY OF STOCK PRICE PREDICTION:

while the 1-day stock price prediction follow closely
with actual stock prices, the predictions for stock prices after 10 days
deviate considerably from the actual stock prices. This shows that machine
learning models fail to provide accurate stock price predictions to retail
investors.

7.2 DEMOCRATIZATION OF MACHINE LEARNING:

An other factor when evaluating the project’s success is when there
investors can use and understand the predictive information provided by the
machine learning technologies using our mobile application . In spite of the
confusions found in some parts of the user interface ,especially in the
advanced user mode, users found useful insights provided by the machine
learning models, such as identifying stocks with upside potential .The result
is significant, in the sense that users with little background on machine
learning technology and stock trading could find potential use cases for the
application .The result simply that machine learning technologies could be
democratized to serve the interest of the general public. Stock price
prediction is a particularly exciting area, because the level of expertise
required to succeed in making profitable short-term investments is
considered to be prohibitive for small , retail investors , and trading with help
of machine learning is a feat only institutional investors could perform. The
application demonstrates one possible way retail investors could use
machine learning technologies on their own.

43
8 .CONCLUSION AND FUTUER WORK:

The project analysis the foundation for democratizing machine learning

technologies for retail investors, connecting predictions made by machine
learning model store tail investors through a mobile application. It helps
investors navigate through the stock markets with additional analysis and
help them make more informed decisions. The findings demonstrated that
the application provides significance in trend prediction. When compared to
the baseline, the prediction shows useful trend tendency with the real stock
trend. Through the application interface, the user can easily compare the
predictions and model scores from different machine learning models, then
choosing the one that fits their preference. The models used in the
application will continue to improve itself by searching for a better model
topology, structure and hyper parameters through evolution algorithm. The
findings concluded the usefulness of evolution algorithm in lowering the
mean squared error when predicting stock prices, which is helpful for
improving the trend prediction for retail investors. Therefore, with the
application and research findings, to large extent the project team achieved
the aim of creating an user-friendly system for retail investors whom does
not have previous technical knowledge to navigate the machine model
predictions result with useful benchmarks.
There are 4 possible further improvements building upon the findings of this
project. First, multiple approach test of raming the problems could be
explained ,such as Predicting Whether the stock price goes upon down
(binary classification

44
based on the previous stock prices. Other features could being corporated,
such as market news and sentiment .Combined with the development of more
advanced machine learning techniques, the accuracy of the information
provided to retail investors might be improved significantly. Second, a larger
scale of evolution with larger population size and more iterations could also
be tested for achieving better results. Model inputs can also be included into
the evolution algorithm as a variable to optimize. Regularized evolution [38]
can be tested to eliminate old models regardless of their accuracy, which
could allow the algorithm to search for more distant models in the search
space. Third, it is also possible to use more finance-specific scores, like those
introduced, as the objective function instead of simple mean squared errors
to achieve better results. Fourth, mobile applications with better presentation
of stock price predictions could be developed to help investors understand the
implications of the stock price predictions, e.g. when to buy or sell. This would
allow investors to make more informed decisions based on the machine
learning models and truly democratize machine learning technologies, which
were believed to be only in the hands of very few people.

45
 REFERNCE

[1] “Survey Finds Hong Kong Securities Market Attracts Wide Range of
Investors,” HKEX, 13
Jul2017;http://www.hkex.com.hk/news/news- release/2017/170713news?
sc_lang=en. Y. Dai and Y. Zhang, “Machine Learning in Stock Price Trend
Forecasting,” Stan ford University;
http://cs229.stanford.edu/proj2013/DaiZhang-
MachineLearningInStockPriceTrendForecasting.pdf.
[2]J. Patel, S. Shah, P. Thakkar, and K. Kotecha, “Predicting stock and
stock price index movement using Trend Deterministic Data Preparation
and machine learning techniques, ”Expert Systems with Applications: An
International Journal, Vol. 42, Jan. 2015, pp. 259-268
[3]B. Wanjawa and L. Muchemi, “ANN Model to Predict Stock Prices at
Stock Exchange Markets,” arXiv:1502.06434 [q-fin.ST], 2014
[4] D. Mandic and J. Chambers, Recurrent Neural Networks for Prediction, Wiley, 2001
[5] R. Williams and D. Zipser, “Gradient-based learning algorithms for
recurrent networksandtheircomputationalcomplexity”,inBack-
propagation:Theory,ArchitecturesandAppli-
Cations, Hillsdale, NJ: Erlbaum, 1992, pp. 433 - 486
[6]S. Hochreiter, Y. Bengio, P. Frasconi, J. Schmidhuber, “Gradient flow in
recurrent nets: the difficulty of learning long-term dependencies”, in A Field
Guide to Dynamical Recurrent Neural Networks, S. C. Kremer and J. F. Kolen, eds.,
IEEE press, 2001
[7] S. Hochreiter and J. Schmidhuber, “Long Short-term Memory”,Neural
Computation, vol. 9,no. 8, pp. 1735 - 1780, 1997
[8]K. Cho et al., “Learning Phrase Representations using RNN Encoder-
Decoder for Statistical Machine Translation” , arXiv:1406. 1078 [cs.CL],
2014.
[9]W. Gail, G. Yoav, and Y. Eran, "On the Practical Computational Power of
Finite PrecisionRNNs for Language Recognition", arXiv:1805.04908
46
[cs.NE], 2018
[10] J. Chung, C. Gulcehre, K. Cho and Y. Bengio, "Empirical Evaluation
of Gated RecurrentNeural Networks on Sequence Modeling".
arXiv:1412.3555 [cs.NE]. 2014
[11] E. Real, et al. “Large-Scale Evolution of Image Classifiers,”arXiv:1703.01041

47
 [cs.NE]. Jun2017.
[12]D. Alajbeg, Z. Bubas and D. Vasic, “Price Distance To Moving Averages
And SubsequentReturns”, International Journal of Economics, Commerce and
Management, Vol. V, Dec 2017,pp. 33 - 47
[13] Progressive Web Apps, Google.
Available:https://developers.google.com/web/progr
essive- web-apps/
[14] Neoteric, “Single-page
application vs. multiple-page application”,
Medium. 2016.
Available:https://medium.com/@NeotericEU/single-
page- application-vs-multiple-page-application-
2591588efe58
[15] Keras: The Python Deep Learning library, Keras. Available:https://keras.io/.

[16] Documentation of scikit-learn

0.19.2. Available:http://scikit-
learn.org/stable/documentation.html.
[17] Alpha Vantage API Documentation, Alpha
Vantage.
Availablehttps://www.alphavantage.co/documenT/.
NumPy v1.14 Manual, The SciPy community.
Available: https://docs.scipy.org/doc/numpy-1.14.5/
[18] pandas: powerful Python data analysis
toolkit, Pandas.
Available:http://pandas.pydata.org/pandas-
docs/version/0.23/.
[20] Hello, Colaboratory - Colaboratory
- Google, Google.
Available :https://colab.research.google.com.
[21] Tensor Board: Visualizing Learning, Google.
48
Available:https://www.tensorflow.org/guide/summaries_a
nd_ten sorboard.
[22] Welcome to Flask — Flask 1.0.2 documentation, Pallets
Team. Available :http://flask.pocoo.org/docs/1.0/.

49
[23] Cloud Storage, Google. Available: https://firebase.google.com/docs/storage/.
[24] Getting Started, The Investors
Exchange. Available
:https://iextrading.com/developer/docs/.
[25] Cloud Functions for Firebase,
Google.
Available:https://firebase.google.com/docs/
funct on/.
[26] Cloud Fire store, Google. Available: https://firebase.google.com/docs/firestore/.
[27]React – A JavaScript library for building user interfaces, Facebook
Inc;https://reactjs.org/.
[28] React Router: Declarative Routing for React.js, React
Training.
Available:https://reacttraining.com/react-router/web/guides/phil
osophy.
[29] Read Me - Redux, Redux. Available:
https://redux.js.org/. [30]Immutable collections for
JavaScript, Facebook Inc. Available:https://github.com/
face book/immutable-j s/.
[31] Introduction - Material Design, Google. Available
:https://material.io/design/introduction/.
[32] Material UI, Material UI Team. Available :https://material-ui.com/

50
 APPENDIX
A. SOURCE CODE:

Predict stock:
import os
import sys
import
requests
import numpy as np
from keras.models import
Sequential from keras.layers
import Dense

FILE_NAME = 'historical.csv'

def get_historical(quote):
# Download our file from google finance
url =
'http://www.google.com/finance/historical?q=NASDAQ
%3A'+quote+'&output=csv'
r = requests.get(url, stream=True)

if r.status_code != 400:
with open(FILE_NAME, 'wb')
as f: for chunk in r:
f.write(chunk)

return True

def stock_prediction():

51
# Collect data points from csv
dataset = []

52
 with open(FILE_NAME) as
f: for n, line in
enumerate(f):
if n != 0:
str =
line.split(',')[1] if
str != "-":
dataset.append(float(line.split(',')

[1])) dataset = np.array(dataset)

# Create dataset matrix (X=t and

Y=t+1) def
create_dataset(dataset):
dataX = [dataset[n+1] for n in
range(len(dataset)2)] return
np.array(dataX), dataset[2:]

trainX, trainY = create_dataset(dataset)

# Create and fit Multilinear Perceptron

model model = Sequential()
model.add(Dense(8, input_dim=1,
activation='relu')) model.add(Dense(1))

model.compile(loss='mean_squared_error',opti
mizer= adam')
model.fit(trainX, trainY, nb_epoch=200,
batch_size=2, verbose=2)

53
 # Our prediction for tomorrow
prediction =
model.predict(np.array([dataset[0]])) result
= 'The price will move from %s to %s' %
(dataset[0], prediction[0][0])

54
 return result

# Ask user for a stock quote

stock_ quote = input('Enter a stock
quote from NASDAQ (e.j: AAPL, FB,
GOOGL): ').upper()

# Check if we got the historical

data if not get_
historical(stock_ quote):
print ('Google returned a 404, please re-run
the script and')
print ('enter a valid stock quote from
NASDAQ') sys.exit()

# We have our file so we create the neural net

and get the prediction
print (stock _prediction())

# We are done so we delete the csv file

os.remove (FILE_NAME)

55
Prophet:
import pandas as
pd import numpy
as np
import pandas_datareader.data as
web from fbprophet import Prophet
import datetime
from flask import Flask,
render_template from flask import
request, redirect
from pathlib import Path
import os
import
os.path
import csv
from itertools import

zip_longest app = Flask(

name )

@app.after_request
def
add_header(response
): """
Add headers to both force latest IE rendering engine or Chrome
Frame, and also to cache the rendered page for 10 minutes.
"""
response.headers['X-UA-Compatible'] =
'IE=Edge,chrome=1' response.headers['Cache-
Control'] = 'public, max-age=0' return response
56
@app.route("/
") def
first_page():
"""
original_end = 175
forecast_start = 200

57
 stock = "IBM"
return render_template("plot.html", original = original_end, forecast =
forecast_start, stock_tinker = stock)
"""
tmp = Path("static/prophet.png")
tmp_csv =
Path("static/numbers.csv") if
tmp.is_file():
os.remove(tm
p) if
tmp_csv.is_file():
os.remove(tmp_csv)
return render_template("index.html")

#function to get stock data

def yahoo_stocks(symbol, start, end):
return web.DataReader(symbol, 'yahoo', start, end)

def get_historical_stock_price(stock):
print ("Getting historical stock prices for stock ", stock)

#get 7 year stock data for Apple

startDate =
datetime.datetime(2010, 1, 4)
#date =
datetime.datetime.now().date()
#endDate = pd.to_datetime(date)
endDate = datetime.datetime(2017, 11, 28)
stockData = yahoo_stocks(stock, startDate,
endDate) return stockData

58
@app.route("/plot" , methods = ['POST',
'GET'] ) def main():
if request.method == 'POST':
stock = request.form['companyname']

59
 df_whole =

get_historical_stock_price(stock) df =

df_whole.filter(['Close'])

df['ds'] = df.index
#log transform the ‘Close’ variable to convert non-stationary data to
stationary. df['y'] = np.log(df['Close'])
original_end = df['Close'][-1]

model = Prophet()
model.fit(df)

#num_days = int(input("Enter no of days to predict stock price for: "))

num_days = 10
future =
model.make_future_dataframe(periods=num_days)
forecast = model.predict(future)

print (forecast[['ds', 'yhat', 'yhat_lower', 'yhat_upper']].tail())

#Prophet plots the observed values of our time series (the black dots),
the forecasted values (blue line) and
#the uncertainty intervalsof our forecasts (the blue shaded regions).

#forecast_plot =
model.plot(forecast)
#forecast_plot.show()

60
#make the vizualization a little better to
understand df.set_index('ds', inplace=True)
forecast.set_index('ds', inplace=True)
#date = df['ds'].tail(plot_num)

61
 viz_df = df.join(forecast[['yhat', 'yhat_lower','yhat_upper']],
how = 'outer') viz_df['yhat_scaled'] = np.exp(viz_df['yhat'])

#close_data = viz_df.Close.tail(plot_num)
#forecasted_data =
viz_df.yhat_scaled.tail(plot_num) #date =
future['ds'].tail(num_days+plot_num)

close_data = viz_df.Close
forecasted_data =
viz_df.yhat_scaled date =
future['ds']
#date = viz_df.index[-plot_num:-1]
forecast_start = forecasted_data[-
num_days]

d = [date, close_data, forecasted_data]

export_data = zip_longest(*d, fillvalue = '')
with open('static/numbers.csv', 'w', encoding="ISO-8859-1", newline='')
as myfile: wr = csv.writer(myfile)
wr.writerow(("Date", "Actual",
"Forecasted"))
wr.writerows(export_data)
myfile.close()

return render_template("plot.html", original = round(original_end,2), forecast =

round(forecast_start,2), stock_tinker = stock.upper())
'''
if name == " main
": main()
62
'''

if name == " main ":

app.run(debug=True,
threaded=True

63
prophet without flask:
import pandas as
pd import numpy
as np
import pandas_datareader.data as
web from fbprophet import Prophet
import datetime
import matplotlib.pyplot as plt

plt.rcParams['figure.figsize']=(20,10)
plt.style.use('ggplot')

#function to get stock data

def yahoo_stocks(symbol, start, end):
return web.DataReader(symbol, 'yahoo', start, end)

def get_historical_stock_price(stock):
print ("Getting historical stock prices for stock ", stock)

#get 7 year stock data for Apple

startDate =
datetime.datetime(2010, 1, 4) #
date =
datetime.datetime.now().date()
# endDate = pd.to_datetime(date)
endDate = datetime.datetime(2017, 11, 27)
stockData = yahoo_stocks(stock, startDate,
64
 endDate) return stockData

def main():

65
 stock = input("Enter stock name(ex:GOOGL, AAPL):
") df_whole = get_historical_stock_price(stock)

df = df_whole.filter(['Close'])

df['ds'] = df.index
#log transform the ‘Close’ variable to convert non-stationary data to
stationary. df['y'] = np.log(df['Close'])

model = Prophet()
model.fit(df)

num_days = int(input("Enter no of days to predict stock price for: "))

future =
model.make_future_dataframe(periods=num_days)
forecast = model.predict(future)

print (forecast[['ds', 'yhat', 'yhat_lower', 'yhat_upper']].tail())

#Prophet plots the observed values of our time series (the black dots), the
forecasted values (blue line) and
#the uncertainty intervalsof our forecasts (the blue shaded
regions). forecast_plot = model.plot(forecast)
forecast_plot.show()

#make the vizualization a little better to

understand df.set_index('ds', inplace=True)
forecast.set_index('ds', inplace=True)

viz_df = df.join(forecast[['yhat', 'yhat_lower','yhat_upper']], how = 'outer')

66
viz_df['yhat_scaled'] = np.exp(viz_df['yhat'])

67
 fig = plt.figure()
ax1 =
fig.add_subplot(111) #
ax1.xaxis_date()
ax1.plot(viz_df.index, viz_df.Close)
ax1.plot(viz_df.index, viz_df.yhat_scaled,
linestyle=':')
ax1.set_title('Actual Close (Orange) vs Close Forecast (Black)')
ax1.set_ylabel('Closing Price in Dollars')
ax1.set_xlabel('Date')

L = ax1.legend() #get the legend

L.get_texts()[0].set_text('Actual Close') #change the legend text for 1st plot
L.get_texts()[1].set_text('Forecasted Close') #change the legend text for 2nd plot

plt.savefig('graph/prophet.png',
bbox_inches='tight') plt.show()

#plot using dataframe's plot function

viz_df['Actual Close'] = viz_df['Close']
viz_df['Forecasted Close'] =
viz_df['yhat_scaled']

viz_df[['Actual Close', 'Forecasted Close']].plot()

if name == " main

": main()

68
INDEX HTML:

<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<title> Stock Value Prediction </title>
<link rel="stylesheet" href="../static/style.css">
</head>

<video plays in line auto play muted loop id="bgvid">

<source src="{{ url _for('static', filename='video/video.mp4') }}"
type="video/mp4">
</video>
<body>
<div >
<form action=http://localhost:5000/plot method = "POST">
<input type="text" name="company name" placeholder="Enter
company ticker symbol" class = "search box">
<input type="submit" name="Search" class="submit">
</form>
<a href="/" class = "homepage">Stock Value Prediction</a>
<marquee class = "bottom bar">Stock values are subjected to
market risk. Prediction is based on machine learning models. Please invest at
your own risk.</marquee>
</div>
</body>

</html>

HTML PLOT:

<html>
<head>
<meta charset="utf-8">
<title> Prediction Results </title>
<script src="/static/js/dygraph.js"></script>
<link rel="stylesheet" href="../static/style_result.css">
<link rel="stylesheet" href="/static/css/dygraph.css"/>
</head>

<body class = "bodyclass" >

<div class="topnav" >

<a href="/" class = "homepage" style="color:White" >Stock Value Prediction</a>
</div>
69
 <h6>
<object class = "summary" data="../static/summary/{{stock_tinker}}.txt"
style="width:85%; height:40%;" ></object>
</h6>

<table class="print_details" align="center" >

<tr>

<td>
<font color="grey">{{ stock_tinker }}</font>
</td>

<td>
<font color="grey">Last Closing Stock Value</font>
</td>

<td>
<font color="grey">Predicted Stock Value</font>
</td>

</tr>

<tr>
<td></td>

<td>
<font color="grey">{{original}}</font>
</td>

<td>
<script type="text/javascript">

if ({{forecast}} > {{original}})

document.write("<center><b><font

color=\"green\"><font size=\"5\"> {{forecast}}

</font><\/font></b></center>");
else
document.write("<center><b><font

color=\"red\"><font size=\"5\"> {{forecast}}

</font><\/font></b></center>");
//<font color="grey">{{forecast}}</font>

</script>
70
 </td>

</tr>

71
 </table>

<script type="text/javascript">
if( {{original}} - {{forecast}} < 0)
{
document.write("<center><b><font

color=\"green\"><font size=\"5\">BUY</font><\/font></b></center>");

}
else
{
document.write("<center><b><font color=\"red\"><font
size=\"5\">SELL - if stock is owned </font><\/font></b></center>");
//document.write("<center><b><font
color=\"red\">Sell - if
stock is
owned<\/font></b></center>");
//document.write("<center><b><font color=\"green\">Short - if stock is
not owned<\/font></b></center>")
document.write("<center><b><font color=\"green\"><font
size=\"5\">SHORT - if stock is not owned</font><\/font></b></center>");
}
</script>

<table class="container">
<tr>
<td>
<div id="graph" style="width:1000px; height:500px;" class="graph"></div>
</td>
<td>
<div id="graphLegend" style="width:20px; height:3px;"
class="graphbox"></div>
</td>
</tr>
<tr>
<td>
<div class =
"labels"> Show
Data:
<input type=checkbox id="0" checked onClick="change(this)">
<label for="0"> <font color="008000">Actual</font></label>
<input type=checkbox id="1" checked onClick="change(this)">
<label for="1"> <font color="#000080">Forecasted</font></label>
</div>
</td>
72
 </tr>
</table>

<script
type="text/javascript"
> g2 = new Dygraph(

73
 document.getElementById("graph"),
// CSV or path to a CSV file.
"{{ url_for('static', filename='numbers.csv') }}", // path to CSV file
{
visibility: [true, true],
labelsDiv: document.getElementById("graphLegend")
} // options
);
setStatus();

function setStatus() {
document.getElementById("visibility").inner
HTML = g2.visibility().toString();
}
function change(el) {
g2.setVisibility(parseInt(el.id),
el.checked); setStatus();
}
</script>

<div><marquee>
<table class="newsbox" scrollamount="1" >
<tr>
<td>
<div>
<a style="display:block;color:black; width: 200px; margin-left:
20px" font- family: "Comic Sans MS"
href="https://www.cnbc.com/2017/12/03/bitcoin-hits-all-time- high-above-
11700-as-recovery-accelerates.html" target="_blank"
><font
color="808080"><b>Bitcoin hits all-time high above $11,700 as recovery
accelerates</font color></b></a>
</div>
</td>
<td>
<div>
<a style="display:block;color:black; width: 200px; margin-
left: 20px" href="http://beta.latimes.com/business/la-fi-markets-20171129-
story.html" target="_blank" ><b><font color="808080">Tech stocks drop
steeply, offsetting bank and healthcare shares' rise</font color></b></a>
</div>
</td>
<td>
<div>
<a style="display:block;color:black; width: 200px; margin-
left: 20px" href="https://www.cnbc.com/2017/11/30/amazon-gets-its-most-
74
bullish-call-yet-1500- price-target.html" target="_blank" ><b><font
color="808080">Wall Street is giddy over Amazon shares, gets its most
bullish call yet: $1,500 price target</font color></b></a>

75
 </div>
</td>
<td>
<div>
<a style="display:block;color:black; width: 200px; margin-left: 20px"
href="http://fortune.com/2017/03/31/amazon-stock-trillion-dollar-company-
apple-tesla- google/" target="_blank" ><b><font color="808080">Amazon
and the Race to Be the First
$1 Trillion Company</font color></b></a>
</div>

</td>
<td>
<div>
<a style="display:block;color:black; width: 200px; margin-left: 20px"
href="https://www.forbes.com/sites/timworstall/2014/04/13/fun-number-
apple-has-twice- as-much-cash-as-the-us-government/#790e34d75570"
target="_blank" ><b><font color="808080">Apple Has Twice As Much
Cash As The US Government</font color></b></a>
</div>

</td>
</tr>
</table>
</marquee>
<marquee direction = "right">
<table class="newsbox" scrollamount="1">
<tr>
<td>
<div>
<a style="display:block;color:black; width: 200px; margin-
left: 20px" href="http://www.latimes.com/business/hiltzik/la-fi-hiltzik-cvs-
aetna-20171204- story.html" target="_blank" ><b><font
color="808080">CVS and Aetna say their huge merger will be great for
consumers. Here's why you should be skeptical</font color></b></a>
</div>
</td>
<td>
<div>
<a style="display:block;color:black; width: 200px; margin-
left: 20px" href="https://www.cnbc.com/2017/12/01/amid-tax-bill-euphoria-
one-big-part-of-the- stock-market-is-getting-left-behind-tech.html"
target="_blank"
><b><font
color="808080">Amid tax-bill euphoria, one big part of the stock market is
getting left behind: Tech</font color></b></a>
76
 </div>
</td>
<td>

77
 <div>
<a style="display:block;color:black; width: 200px; margin-left: 20px"
href="https://www.nytimes.com/2017/12/04/business/dealbook/broadcom-
qualcomm.html?rref=collection%2Fsectioncollection
%2Fbusiness&action=click&content
Collection=business&region=rank&module=package&version=highlights&co
ntentPlace ment=5&pgtype=sectionfront" target="_blank" ><b><font
color="808080">Broadcom Puts Qualcomm on
the Clock</font color></b></a>
</div>
</td>
<td>
<div>
<a style="display:block;color:black; width: 200px; margin-
left: 20px" href="https://www.reuters.com/article/us-usa-stocks/dow-hits-
record-as-investors-bet- on-tax-cuts-idUSKBN1DY1EP" target="_blank"
><b><font color="808080">Dow hits record as investors bet on tax
cuts</font color></b></a>
</div>
</td>
<td>
<div>
<a style="display:block;color:black; width: 200px; margin-
left: 20px" href="http://www.businessinsider.com/meet-the-paypal-mafia-the-
richest-group-of-men- in-silicon-valley-2014-9" target="_blank" ><b><font
color="808080">Meet The PayPal Mafia, The Richest Group Of Men In
Silicon Valley</font color></b></a>
</div>
</td>
</tr>
</table>
</marquee>
</div>

</body>
</html>

78
B .OUTPUT:

B.1 SERVING FOR STOCK MARKET PREDICTION

B.2 RUNNING ON HTTP

79
 B.3 OPEN THE LINK IN THE BROWSER

B.4 HOME PAGE OR LOGIN PAGE

80
B.5 OUTPUT OF THE GRAPH

81