STOCK PRICE PREDICTION USING TIME SERIES

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 06 | June 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1194
STOCK PRICE PREDICTION USING TIME SERIES
Mr.P.Ramu1, Ms.A.Vani Priya2, Ms.D.Roopa Sree 3, Ms.S.Sankeerthana4
1Assistant Professor, Dept of Computer Science and Engineering from Sreenidhi Engineering College, JNTUH.,
(T.S.), INDIA.
234 UG Scholar, Dept of Computer Science and Engineering from Sreenidhi Engineering College, JNTUH.,(T.S.),
INDIA.
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Analysts have found it challenging to
estimate a company's stock price because of its volatility
and shifting nature. Because stock values are time-
dependent, this research aims to forecast stock values
using the technique called Time series, which requires
tracking many changes in a single variable over time and is
particularly suited for financial forecasting. Using Time
Series on a dataset will allow you to examine how a
defined economic, currency, or meteorological variable
changes over time, as well as how it changes in
comparison to other similar variables over the same
period. In this project, we will use time series models to
forecast stock values using ARIMA and other forecasting
approaches. Time Series is a basic statistical tool for
analyzing continually changing variables such as stock
prices, weather, currencies, and so on. A popular
forecasting model called the ARIMA model that works
with historical data to provide near-term projections and
may be used as a foundation for more complex and
complicated models. We'll gather stock market data and
analyze it with ARIMA time series modeling and other
forecasting techniques like Naive Estimate and
Exponential Smoothing, to forecast future stock prices.
Key Words: Stock price, time changing, Data, ARIMA,
Exponential smoothing, Naive, Seasonal Naive, statistics,
Analysis
1 INTRODUCTION
This work is about the prediction of the stock price using
time series. Every investor, whether an individual or a
company, wants a good or reasonable return on their
investment. Stocks are one of the best ways to get a good
return on investment. This requires investors to fully
understand many stocks and their current prices. To
maximize profits and avoid losses, you need to make
accurate price forecasts when buying and selling stocks.
Both the Efficient Market Hypothesis and the Elliott Wave
Theory test several predictive principles. The behavior of
institutional investors, often known as large buyers and
sellers, is generally a major contributor to equity value. If
one day there are more buyers than sellers, the auction
will be higher for that price. Finally, the price is displayed
at the control point. This is the average price or the most
constant price. Pricing is usually distributed in most cases.
Therefore, you need to select entry and exit points based
on the auction price to maximize profits and accurately
predict stop-loss points for complete risk analysis.
Extensive statistical techniques such as autoregressive and
moving averages are often used to achieve the same goal.
With the latest computing technologies such as machine
learning, ARIMA (autoregressive integrated moving
average), exponential smoothing, autoregressive
integrated moving, ARIMA (autoregressive integrated
moving average), naive prediction, seasonal naive
prediction, and neural networks several techniques such
as are possible. The currently proposed model uses all
new techniques to predict current stock prices and
maximize profits. Each model is ranked to help users
decide whether to buy or sell a particular stock, whether
the transaction is short-term or long-term. Unlike the old
approach, this model uses all the latest methods and is
more likely to make accurate predictions.
1.1 RELATED WORK
From the literature survey, it had been observed that the
appliance of machine learning techniques to securities
market prediction is being undertaken thoroughly
throughout the globe. Machine Learning techniques are
proving to be rather more accurate and faster as
compared to contemporary prediction techniques.
Significant work has been done throughout the planet in
this field.
Authors Naresh Kumar, and Seba Susan the objective of
this think is to supply an assessment of forecast models
based on COVID-19 cases, as well as to estimate the virus's
effect in influenced nations and around the world [5]. On
COVID-19 occurrences, demonstrate execution was
assessed utilizing measurements such as cruel supreme
mistake (MAE), root cruel square blunder (RMSE), root
relative squared mistake (RRSE), and cruel supreme rate
mistake (MAPE). For COVID-19 affirmed, dynamic,
recuperated, and passing cases, we produce estimating
discoveries. ARIMA outflanked the Prophet's show,
concurring with the findings. COVID-19 occasion day-level
information has been assembled from a GitHub store. The
ESRI living chart book group, the Connected Material

science Lab (APL), and the Center for Frameworks Science
and Building (CSSE) at Johns Hopkins College, both of
which are based within the Joined together States, bolster
and keep up the asset [5]. Starting January 22, 2020, the
store will contain worldwide COVID-19 detailed
occurrences on an everyday premise.
1.2 METHODOLOGY
Stock price prediction is a big problem because it involves
many factors that have yet to be addressed and it doesn’t
seem statistical initially. But by using accurate machine
learning techniques, one can relate previous data to this
data and train the machine to find out from it and make
appropriate assumptions.
In the existing system stock showcase is one of a
country's most imperative financial divisions. It gives
financial specialists the chance to contribute and benefit
from their cash. Analysts from an assortment of spaces,
counting measurements, fake insights, financial matters,
and funds, are all fascinated by anticipating the stock
advertisement. Stock showcase determining precision
brings down showcase risk[12].
When it comes to the stock market's consistency,
there are numerous diverse perspectives. Concurring to
the effective showcase theory (EMH), all open data is
immediately completely coordinated into the current
showcase cost, causing stock cost volatility. Many
machine learning strategies have been utilized within
the writing to assess stock cost heading. A few of these
works are altogether inspected. Ampomah et al. (2020)
explored the execution of tree-based AdaBoost gathering
ML models in determining stock costs (specifically,
AdaBoost-DecisionTree (Ada-DT), AdaBoost-
RandomForest (Ada-RF), AdaBoost-Bagging (Ada-BAG),
and Bagging (Ada-BAG), and Bagging-ExtraTrees (Bag-
ET) [12]. The AdaBoost-ExtraTree (Ada-ET) model
outperformed the other tree-based AdaBoost ensemble
models, according to the findings.
Machine learning strategies such as direct
discriminant examination, arbitrary woodland,
manufactured neural arrange, SVM, and logit were
utilized by the analysts Ernest Kwame Ampomah,
Gabriel Nyame, and Zhiquang qin. The exploratory
discoveries appeared that SVM beat all of the other
classification techniques.
1.3 EXISTING METHODOLOGY
Flow chart -1:Existing method
In the proposed system in this time-series study, the entire
cost of a face drilling rig utilized in the Swedish mining
sector is estimated using an Autoregressive Integrated
Moving Average (ARIMA) model [15].
Time series forecasting forecasts future data points based
on data gathered over a specific period. Forecasted data
points will serve as a foundation for production
management and planning, as well as to optimize
industrial processes and economic planning. The primary
aim is to obtain the best prediction possible, which entails
reducing the mean square difference between actual and
anticipated values for each lead-time.
Time series forecasting approaches such as Box–Jenkins
and the Autoregressive Integrated Moving Average
(ARIMA) are based on the assumption that time series
data is generated by linear processes. Some of the
techniques were used. The authors Al-Douri, and Jan
Lundberg used Multiple regression and neural network
techniques to model and anticipate the future.
The data in a stable stochastic model have the same
variance and autocorrelation. The difficulty in determining
the parameters is the model's flaw. To address this issue
Historical raw data
collection
Data cleaning
Importing libraries
Feature extraction
&scaling
Model
Prediction and
evaluation
Train data
Test data

and provide accurate forecasting, automated model
selection processes are required. Zhang proposes
combining ARIMA and Artificial Neural Network (ANN)
models in a hybrid technique. The combination increases
predicting precision. The results from three real-world
data sets show that the hybrid model outperforms each
component model. The ARIMA and ANN models share
some commonalities. Both have a diverse range of models
with varying model ordering. To create an effective model,
both require a big sample size.
ARIMA, on the other hand, can deliver results based on the
problem and data. The main portion of the ARIMA model
is a complex polynomial that combines AR and MA
polynomials. All of the TC data points are subjected to the
ARIMA (p, d, q) model.[15]The mean of the time series
data; p: the number of autoregressive delays; q: the
number of moving average delays AR (autoregressive
coefficients):
MA stands for moving average coefficients.
d: the number of differences produced by the white noise
in the time-series data.
For TC (Z TC), the ARIMA model is stochastically
implemented using default values for p, d, and q. (0,0,0),
(0,0,1), (0,1,1), (1,0,0), (1,0,1), (1,1,1), (2,1,1) (2,0,3). For
each scenario, all of the TC data from the previous 37
months is included [15].
2. ALGORITHMS USED
2.1 ARIMA (Autoregressive integrated moving
average)
ARIMA may be a blend of two calculations: auto relapse
and moving averages, as the title suggests.
Autoregression could be a time arrangement
demonstrate that employments past time step data as
input to a relapse condition to anticipate values in the
following time step[3]. It could be a clear strategy that
can make solid forecasts for a wide extent of time
arrangement issues. A moving normal could be time-
arrangement information normal that advances through
all the arrangements by subtracting the best things from
the already found the middle value of gather and
embeddings the another in each average.
The Arima model, in some cases known as the Box-
Jenkins model, was presented by George Box and
Gwilym Jenkins[22]. The ARIMA model, which contains
the condition underneath, combines the autoregression
and moving average models (3).
c+ϕ1y′(t−1)+⋯+ϕpy′(t)−p+θ1ε(t−1)+⋯+θqε(t)−q+ε(t)
= y′(t) (3)
On the one hand, we have indicators with lagged y(t)
values and lagged errors, while y′t could be a
subordinate variable that can be shifted a few times. The
ARIMA (p, d, and q) demonstrate is the title given to this
show. The Auto regression and Moving Average
component orders are p and q, individually, and the
degree of differentiation is d.
Graph 1: Arima results
2.2 Exponential smoothing
Exponential smoothing is a time series forecasting
method for univariate data and can be extended to
support data with systematic trends or seasonal
components. This is a powerful predictive method that
can be used as an alternative to Box Jenkins' popular
ARIMA method family.
This determining method includes allotting weights
to earlier information in such a way that they rot
exponentially over time. The foremost later weights are
on best, and as the time figure increments, they start to
debase [3].
Arima is the autoregressive integrated moving average
used for calculating moving averages.
Graph 2:Results of exponential smoothing

2.3 Naive
The estimate is made by applying the taking after equation
(1) to past information without making any expectations.
y (T) = y′ (T + h) |T - (1)
The prior data prediction is (T + h), while the current
data forecast is (T).
Graph 3:Results from Naive
2.4 Seasonal Naive
This estimating strategy is comparative to naive
forecasting but the estimate is based on the previous
information of the same season. The equation is as
follows
y’(T + h) | T =y(T + h) -m(k + 1) -(2)
2.5 Neural Networks
NNAR (p, x) is a nonlinear and advanced forecasting
scenario in which p is the no. of lagged inputs and x is
the no. of hidden layers. It illustrates the architecture
of neural networks.
3 FUTURE SCOPE
The no. of stock cost expectation calculations will be
extended in future investigations. The taking after is a few
cases of how the comes about of this think could be
utilized to figure stock costs: To determine whether
there's any drift or regularity within the information, the
figure of each bank's stock cost must be tried on greater
preparing datasets.
A set of conventional statistics and neural organized
calculations based on slant or regularity must be built to
decide the finest strategy for stock cost prediction. Any
procedure's execution must be assessed utilizing the back-
testing approach.
For a trade user's comfort, the assessed blunder terms
can be spoken to as RMSE. The finest calculations for each
stock cost can be built up due to the least RMSE esteem;
these calculations ought to be utilized to estimate stock
costs, and successful stock cost determination can result in
impressive benefits.
4 CONCLUSION
The ARIMA demonstration and the EXPONENTIAL
SMOOTHING show for stock cost expectations were given
in this research. Each calculation distinguishes the stock
information set of all five educates, concurring with the
assessments of these two models. The ARIMA show test
comes about and appeared that it can dependably
anticipate stock costs within the brief term.
This may lead to advantageous speculation choices for
stock advertising examiners. The ARIMA demonstrate may
be prepared to compete with other short-term forecast
models based on the discoveries obtained. A wide extent
of recurrence values can be utilized utilizing exponential
smoothing. The Exponential smoothing approach was
chosen for a single-time arrangement that was taken after
a design in terms of order choice. There are numerous
well-known time arrangement strategies within the
ARIMA. The plan area of ARIMA was basic, conveying an
about straight line.
The data fed into the system is extracted every month
from Yahoo! Finance, and the data is cleansed by removing
outliers. The time series object is then deconstructed
because proper findings are dependent on several factors.
Following that, the time series objects are supplied to
algorithms like ARIMA, Exponential Smoothing, Nave
Forecasting, Seasonal Nave Forecasting, and Neural
Networks, among others.
While the exactnesses of the other calculations like naïve,
regular naïve, and neural systems are on a normal of
94.7%, ARIMA and Exponential smoothing have given
2.9% more exactness than the rest that's, 97.6% precision
which straightforwardly shows that the mistakes in
ARIMA and Exponential Smoothing are way less
comparatively.
When the information had a solid regular slant, ARIMA
and Exponential smoothing created a reliable
demonstration. In this circumstance, ARIMA and
Exponential smoothing beat other models, be that as it
may, the execution and precision of these two models are
subordinate to the information. We ought to nourish the
information to all of the models, compare the outcomes,
and select the foremost exact that comes about depending
on the rankings.
When the RMSE for each bank's models was compared, it
was found that factual strategies beat the Repetitive
Neural Arrange (RNN) strategy, since the RNN strategy is
way better suited for foreseeing stock advertise returns

than measurable models, which must be adjusted in a case
for stock cost prediction.
REFERENCES
[1] T. Huamin, D. Qiuqun, and X. Shanzhu, "Reconstruction
of time series with missing value using 2D representation-
based denoising autoencoder," in Journal of Systems
Engineering and Electronics, vol. 31, no. 6, pp. 1087-1096,
Dec. 2020, DOI: 10.23919/JSEE.2020.000081.
[2] Ariyo, A. O. Adewumi, and C. K. Ayo, "Stock Price
Prediction Using the ARIMA Model," 2014 UKSim-AMSS
16th International Conference on Computer Modelling and
Simulation, 2014, pp. 106-112, DOI:
10.1109/UKSim.2014.67.
[3] Gupta and A. Kumar, "Mid Term Daily Load Forecasting
using ARIMA, Wavelet-ARIMA, and Machine Learning,"
2020 IEEE International Conference on Environment and
Electrical Engineering and 2020 IEEE Industrial and
Commercial Power Systems Europe (EEEIC/I&CPS
Europe), 2020, pp. 1-5, DOI:
10.1109/EEEIC/ICPSEurope49358.2020.9160563.
[4] International Journal of Innovative Technology and
Exploring Engineering (IJITEE) ISSN: 2278-3075,Volume-9
Issue-5, March 2020. Retrieval Number:
D1869029420/2020©BEIESPDOI:
10.35940/ijitee.D1869.039520

STOCK PRICE PREDICTION USING TIME SERIES

More Related Content

Similar to STOCK PRICE PREDICTION USING TIME SERIES (20)

More from IRJET Journal (20)

Recently uploaded (20)

STOCK PRICE PREDICTION USING TIME SERIES