AMAZON STOCK PRICE PREDICTION BY USING SMLT

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 04 | Apr 2023 www.irjet.net p-ISSN: 2395-0072
© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 1075
AMAZON STOCK PRICE PREDICTION BY USING SMLT
S.M. Ajitha1, S.Karthick2, Mu.Karthikeyan3, N.T.Naveen Balaji4
1 (Assistant Professor) Department of Information Technology, Meenakshi College of Engineering, Chennai, Tamil
Nadu, India
2-4Students of Department of Information Technology, Meenakshi College of Engineering, Chennai, Tamil Nadu,
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Stock prices are determined by a company's
initial public offering (IPO). Investment firms use a variety
of metrics to determine what the stock's price should be.
Traders use financial metrics to determine the value of the
company, including its history of earnings, changes in the
market, and the profit it can reasonably be expected to
bring in. Stock price prediction has become an important
research area. We propose a SMLT algorithms to accurately
predict the stock price. Linear regression algorithm has the
best accuracy with precision, Recall and F1 Score.
Keywords: LINEAR REGRESSION, ACCURACY,
PRECISION
1. INTRODUCTION:
Amazon (AMZN) is one of the most well-known
companies in the world and holds a resolute position. The
stock market prediction problem has drawn a lot of
interest from both scholars and practitioners since it is
crucial but difficult. An RNN-based ensemble model for
financial market forecasting using news releases [1]-[2]
was described in the existing solution. The most
representative elements from financial news and
historical data were culled using sentiment analysis and
the sliding window approach. By using data from the past,
machine learning can forecast the future [3]. Computers
may learn without being explicitly taught thanks to a
technique called machine learning (ML), which is a subset
of artificial intelligence (AI). Algorithms with specialized
functions are used during the training and prediction
processes. The objective is to build a machine learning
model for forecasting the price of Amazon stock [4] that
might ultimately replace the supervised machine learning
regression models, which can be updated by predicting
outcomes with the highest level of accuracy by contrasting
supervised methods [10]. The rest of this article is divided
into the following sections. The three algorithms (KNN,
EN, and SVR) and the structure of our suggested LR
architecture are explained in Section II. In Section III,
experiments and analyses are described. This includes
data preparation, the evaluation index, experimental
outcomes, and statistical analysis. In Section IV,
conclusions are reached.
1.1 Existing System:
When making stock market investing selections, traders
and investors can benefit greatly from financial news
announcements. The stock market prediction problem has
drawn a lot of interest from academics and industry
professionals since it is crucial yet difficult. Due to the
complexity and ambiguity of the natural language [1]
[15][16] used in the news, conventional machine learning
models frequently fail to understand the substance of
financial news. An RNN-based ensemble [4] [5] [6] model
for financial market prediction using news releases was
reported in this study. The sliding window approach and
sentiment analysis were used to extract the most
representative characteristics from historical data and
financial news. Compared to conventional pre-processing
techniques (such bag-of-words and TF-IDF), which extract
tens of thousands of features, this significantly decreased
the number of dimensions.
1.2 Proposed System:
Data gathered from a variety of sources make up the
proposed project. The data will be examined for accuracy.
The cleansed data will be created for testing and training
purposes. The machine learning approach is used to
generate the model. To identify the optimal method, many
algorithms are used. The ideal one serves as a template.
The new stock price is predicted using the data model.
2. ALGORITHM USED:
The Elastic net, KNN, and SVR are three well-known
algorithms capable of handling sequential structural data.
The design of our LR Algorithm, which has the highest
accuracy for stock price forecasts, is then described.
2.1: K-NEAREST NEIGHBORS:
A straightforward technique known as "K nearest
neighbors" stores all of the relevant data and forecasts the
numerical objective based on a similarity metric. (e.g.,
distance functions) [7]-[9]. Since the first decade of the
1970s, KNN has been utilized as a non-parametric method
for statistical estimation and pattern identification. The
supervised machine learning technique known as k-
nearest neighbor’s (KNN) is straightforward and simple to

use. It may be used to tackle classification and regression
issues. The KNN method pre supposes the presence of
nearby comparable objects. Alternatively said, related
objects are [9] close to one another. With some
mathematics we may have learnt in our early years—
calculating the distance between points on a graph—KNN
encapsulates the notion of similarity (also termed
distance, proximity, or closeness).
Fig-1: Predicted accuracy using KNN
2.2: SUPPORT VECTOR REGRESSION:
Machine learning classification and regression issues are
frequently solved with Support Vector Machines (SVM)
[20]-[22]. The classification problem is a generalization of
the regression problem, in which the model produces an
output with continuous values as opposed to one from a
limited set. In other terms, a regression model calculates
the estimate of a multivariate continuous function. SVMs
formulate binary classification issues as convex
optimization issues and use those to solve the issues. The
goal of the optimization issue is to accurately categorize as
many training points as feasible while determining the
largest margin separating the hyperplane. Support vectors
in SVMs are used to represent this ideal hyperplane. The
SVM is adaptable to regression issues due to its sparse
solution and strong generalization. In order to generalize
SVM to SVR, the function is surrounded by a -insensitive
zone known as the supervised learning context.
Fig-2: Predicted accuracy using SVR
2.3: ELASTIC NET:
A model that assumes a linear connection between the
input variables and the target variable is known as a
"linear regression" model. This connection may be
conceptualized as a hyperplane connecting the input
variables to the target variable in higher dimensions and
as a line with a single input variable [16]-[19]. An
optimization procedure is used to find the model's
coefficients in order to reduce the total squared error
between the predictions (yhat) and the anticipated goal
values. (y).
Loss = sum i=0 to n (y_i – yhat_i)^2
The predicted coefficients of the model might grow
significantly, which makes the model sensitive to inputs
and potentially unstable. This is especially true for issues
when there are fewer observations (samples) or more
samples (n) than input variables or predictors (p) (so-
called p >> n issues).
A common punishment is to devalue a model according to
the sum of the squared coefficient values. The punishment
for this is an L2. Although it prevents any coefficients from
being dropped from the model, an L2 penalty reduces the
size of all coefficients.
L2_penalty = j=0 to p beta_j2
Another common punishment is to devalue a model
according to the total of the absolute coefficient values. It
is known as the L1 penalty. All coefficient sizes are
reduced by an L1 penalty, which also permits certain
coefficients to be minimized to zero, which eliminates the
predictor from the model.Sum j=0 to p abs
l1_penalty(beta_j)A penalized linear regression model
called an elastic net incorporates both L1 and L2 penalties
during training. The weight assigned to the L1 and L2
penalties is specified by the hyper parameter "alpha,"
which is taken from "The Elements of Statistical Learning."
The L1 penalty's contribution is weighted using an alpha
value between 0 and 1, and the L2 penalty's contribution
is weighted using one minus the alpha value. ((1 - alpha) *
l2_penalty) + (alpha * l1_penalty) =
elastic_net_penalty. For instance, an alpha of 0.5 would
allow each penalty to contribute equally to the loss
function. All weight is given to the L2 penalty with an
alpha value of 0, while a value of 1 yields all weight to the
L1penalty. Elastic net has the advantage of allowing a
balance between the two penalties, which can lead to
better performance on particular problems than a model
with only one penalty. Lambda, a further hyperparameter,
affects how much the total of both penalties is weighted in
the loss function. The fully weighted penalty is applied by
default with a value of 1.0; if the value is zero, the penalty
is not applied. Lambda values as tiny as 1e-3 or less are
typical.
Loss+(lambda*elastic_net_penalty)=elastic_net_loss.

Fig-3: Predicted accuracy using Elastic Net
2.4: LINEAR REGRESSION:
A machine learning algorithm built on supervised
learning is linear regression. The job of predicting the
value of a dependent variable (y) based on an independent
variable is carried out using linear regression. (x). Thus, x
(the input) and y (the output) are shown to be linearly
related using this regression approach.(output). It
completes a regression job. Regression creates a goal
prediction value based on independent variables. It is
mostly used to determine how variables and forecasting
relate to one another. The type of link that different
regression models take into account between the
dependent and independent variables, as well as the
amount of independent variables utilized, vary. The job of
predicting the value of a dependent variable (y) based on
an independent variable is carried out using linear
regression. (x). Consequently, this regression approach
identifies a linear connection between x(input) and
y(output).
The linear regression procedure, often known as
linear regression, illustrates a linear connection between
one or more independent variables (y) and a dependent
variable (y). As a result of displaying a linear connection,
linear regression may be used to determine how the value
of the dependent variable [23]-[25] changes in proportion
to the value of the independent variable. A sloping straight
line illustrating the relationship between the variables is
provided by the linear regression model.
The general form of each type of regression is:
 Simple linear regression: Y = a + bX + u
 Multiple linear regression: Y = a + b1X1
+ b2X2 +
b3X3 + ... + btXt + u
Where:
 Y = the variable that you are trying to predict
(dependent variable).
 X = the variable that you are using to predict Y
(independent variable).
 a = the intercept.
 b = the slope.
 u = the regression residual.
Fig-4: Predicted accuracy using Linear Regression
Fig-5: Architecture of Proposed system
Fig-6: Histogram comparison of different values
2.5: DEPLOYEMENT:
DJANGO:
A Python-based server-side web framework with a wide
range of features, Django, is very well-liked. Django is one
of the most widely used web server frameworks, how to
set up a development environment, and how to start using
it to build your own web applications Regardless of the
popular operating system you use, this article should
provide you with all you need to begin building Django
apps. A high-level Python web framework called Django
enables the quick creation of safe and dependable
websites. Django, which was created by seasoned
programmers, handles a lot of the pain associated with

web development, allowing you to concentrate on
developing your app without having to recreate the wheel.
HTML:
Hyper Text Markup Language is referred to as HTML.
Using a markup language, it is used to create web pages.
Markup language and hypertext are combined to form
HTML. The link between web pages is defined by
hypertext. To specify the text document within the tag that
specifies the structure of web pages, a markup language is
employed. This language is used to annotate (create
computer-readable comments on) material so that a
computer can comprehend it and change the content as
necessary. Most markup languages, including HTML, are
understandable by humans. Tags are used in the language
to specify what text processing is required.
CSS:
Cascading Style Sheets are known as CSS. It is a language
for defining how Web pages are presented, including the
colors, layout, and fonts that make our web sites appealing
to users. For creating style sheets for the web, CSS was
created. Any XML-based markup language may be used
with it because it is not dependent on HTML. Let's attempt
to decode the acronym now:
• Cascading: Falling into Old Habits
• Sheets: Writing in various documents using our style
3. CONCLUSION:
Customers may buy a range of products and services
from Amazon.com, Inc. Through its shops, customers may
purchase goods from both its own third-party sellers and
those it has bought to resell. Additionally, it manufactures
and sells electrical equipment. The project's primary
objectives are to determine accuracy, reduce mistake
rates, and obtain results from deployment. Data cleansing
and processing, missing value analysis, exploratory
analysis, and model creation and assessment came next in
the analytical process. The best accuracy on the public test
set with the highest accuracy score will be discovered. By
employing the algorithm with the highest degree of
accuracy, this can assist in determining the Amazon Stock
Price.
4. FUTURE WORK:
To link the AI Model, forecast the price of Amazon.com.
This procedure may be made automated by displaying the
prediction results in a desktop or online application. To
more efficiently do the job that has to be done in an AI
environment.
REFERENCES
[1] Raymond Chiong , Zongwen Fan, Zhongyi Hu, and
Sandeep Dhakal “A Novel Ensemble Learning
Approach for Stock Market Prediction Based on
Sentiment Analysis and the Sliding Window
Method”,2022.
[2] T. Fischer and C. Krauss, “Deep learning with long
short-term memory 600 networks for financial
market predictions,” Eur. J. Oper. Res., vol. 270, 601
no. 2, pp. 654–669, Dec. 2017. 602
[3] X. Wu, Q. Ye, H. Hong, and Y. Li, “Stock selection
model based on 603 machine learning with wisdom of
experts and crowds,” IEEE Intell. 604 Syst., vol. 35, no.
2, pp. 54–64, Mar. 2020. 605
[4] M. Jiang, L. Jia, Z. Chen, and W. Chen, “The two-stage
machine 606 learning ensemble models for stock
price prediction by combining 607 mode
decomposition, extreme learning machine and
improved har- 608 mony search algorithm,” Ann.
Oper. Res., vol. 309, pp. 553–585, 609 Jun. 2020. 610
[5] D. C. Furlaneto, L. S. Oliveira, D. Menotti, and G. D. C.
Cavalcanti, 611 “Bias effect on predicting market
trends with EMD,” Expert Syst. Appl., 612 vol. 82, pp.
19–26, Oct. 2017. 613
[6] I. E. Fisher, M. R. Garnsey, and M. E. Hughes, “Natural
language 614 processing in accounting, auditing and
finance: A synthesis of the 615 literature with a
roadmap for future research,” Intell. Syst. Accounting,
616 Finance Manage, vol. 23, no. 3, pp. 157–214, Jul.
2016. 617
[7] G. G. Creamer, “Network structure and market risk in
the Euro- 618 pean equity market,” IEEE Syst. J., vol.
12, no. 2, pp. 1090–1098, 619 Jun. 2018. 620
[8] S. Feuerriegel and J. Gordon, “Long-term stock index
forecasting based 621 on text mining of regulatory
disclosures,” Decis. Support Syst., vol. 112, 622 pp. 88–
97, Aug. 2018. 623
[9] T. Geva and J. Zahavi, “Empirical evaluation of an
automated intraday 624 stock recommendation
system incorporating both market data and textual
625 news,” Decis. Support Syst., vol. 57, pp. 212–223,
Jan. 2015. 626
[10] F. Z. Xing, E. Cambria, and R. E. Welsch, “Natural
language based 627 financial forecasting: A survey,”
Artif. Intell. Rev., vol. 50, no. 1, 628 pp. 49–73, Jun.
2018. 629

[11] K. Yuan, G. Liu, J. Wu, and H. Xiong, “Dancing with
Trump in the 630 stock market: A deep information
echoing model,” ACM Trans. Intell. 631 Syst. Technol.,
vol. 11, no. 5, pp. 1–22, Oct. 2020. 632
[12] K. Nam and N. Seong, “Financial news-based stock
movement prediction using causality analysis of
influence in the Korean stock market,” 634 Decis.
Support Syst., vol. 117, pp. 100–112, Feb. 2019. 635
[12] J. Nikkinen and J. Peltomäki, “Crash fears and
stock market effects: 636 Evidence from web searches
and printed news articles,” J. Behav. 637 Finance, vol.
21, no. 2, pp. 117–127, 2019. 638
[13] S. Evert et al., “Combining machine learning and
semantic features in 639 the classification of
corporate disclosures,” J. Log., Lang. Inf., vol. 28, 640
no. 2, pp. 309–330, 2019. 6M. Young, The Technical
Writer’s Handbook. Mill Valley, CA: University Science,
1989.
[14] E. Saquete, D. Tomás, P. Moreda, P. Martínez-Barco,
and M. Palomar, 643 “Fighting post-truth using
natural language processing: A review 644 and open
challenges,” Expert Syst. Appl., vol. 141, Mar. 2020,
645 Art. no. 112943. 646
[15] N. Seong and K. Nam, “Predicting stock movements
based on financial 647 news with segmentation,”
Expert Syst. Appl., vol. 164, Feb. 2021, 648 Art. no.
113988. 649
[16] R. Chiong, G. S. Budhi, S. Dhakal, and F. Chiong, “A
textual-based 650 featuring approach for depression
detection using machine learning 651 classifiers and
social media texts,” Comput. Biol. Med., vol. 135, 652
Aug. 2021, Art. no. 104499. 653
[17] S. F. A. Zaidi, F. M. Awan, M. Lee, H. Woo, and C.-G. Lee,
654 “Applying convolutional neural networks with
different word repre655 sentation techniques to
recommend bug fixers,” IEEE Access, vol. 8, 656 pp.
213729–213747, 2020. 657
[18] S. M. A. Islam, B. J. Heil, C. M. Kearney, and E. J. Baker,
“Protein 658 classification using modified n-grams
and skip-grams,” Bioinformatics, 659 vol. 34, no. 9, pp.
1481–1487, May 2018. 660
[19] S. Yilmaz and S. Toklu, “A deep learning analysis on
question classi661 fication task using Word2vec
representations,” Neural Comput. Appl., 662 vol. 32,
pp. 2909–2928, Jan. 2020. 663
[20] J. Lilleberg, Y. Zhu, and Y. Zhang, “Support vector
machines and 664 Word2vec for text classification
with semantic features,” in Proc. IEEE 665 14th Int.
Conf. Cognit. Informat. Cognit. Comput. (ICCI*CC), Jul.
2015, 666 pp. 136–140. 667
[21] Y. Qiu, H.-Y. Yang, S. Lu, and W. Chen, “A novel hybrid
model based 668 on recurrent neural networks for
stock market timing,” Soft Comput., 669 vol. 24, no. 20,
pp. 15273–15290, 2020. 670
[22] T. Young, D. Hazarika, S. Poria, and E. Cambria, “Recent
trends in 671 deep learning based natural language
processing,” IEEE Comput. Intell. 672 Mag., vol. 13, no.
3, pp. 55–75, Aug. 2018. 673
[23] M. Kraus and S. Feuerriegel, “Decision support from
financial disclo674 sures with deep neural networks
and transfer learning,” Decis. Support 675 Syst., vol.
104, pp. 38–48, Dec. 2017. 676
[24] A. Atkins, M. Niranjan, and E. Gerding, “Financial news
predicts stock 677 market volatility better than close
price,” J. Finance Data Sci., vol. 4, 678 no. 2, pp. 120–
137, Jun. 2018. 679
[25] Y. Liu, Y. Chen, S. Wu, G. Peng, and B. Lv, “Composite
leading search 680 index: A preprocessing method of
internet search data for stock trends 681 prediction,”
Ann. Oper. Res., vol. 234, no. 1, pp. 77–94, Nov. 2015.

AMAZON STOCK PRICE PREDICTION BY USING SMLT

Recommended

More Related Content

Similar to AMAZON STOCK PRICE PREDICTION BY USING SMLT (20)

More from IRJET Journal (20)

Recently uploaded (20)

AMAZON STOCK PRICE PREDICTION BY USING SMLT