Novel approach for predicting the rise and fall of stock index for a specific company 2

INTERNATIONALComputer Engineering and Technology ENGINEERING
International Journal of JOURNAL OF COMPUTER (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME
& TECHNOLOGY (IJCET)

ISSN 0976 – 6367(Print)
ISSN 0976 – 6375(Online) IJCET
Volume 4, Issue 2, March – April (2013), pp. 23-30
© IAEME: www.iaeme.com/ijcet.asp
Journal Impact Factor (2013): 6.1302 (Calculated by GISI)
©IAEME
www.jifactor.com

NOVEL APPROACH FOR PREDICTING THE RISE AND FALL OF
STOCK INDEX FOR A SPECIFIC COMPANY

Anirudh Prabhu1, Aldrin Rodrigues1, Chirag Chauhan1, G.Anuradha2
1
Student, St. Francis Institute of Technology, Borivali (West), Mumbai – 4001031
2
Associate Professor, St. Francis Institute of Technology, Borivali (West),
Mumbai – 4001032

ABSTRACT

The financial market is highly fluctuating in nature. If investment strategies are not
adequately planned and designed, it may lead to high monetary losses. There’s a high risk
involved in stock market investment as the investment strategies are purely based on expert
knowledge about the market conditions and some amounts of instincts.
In our proposed method using the insight of technical analysis and the concept of
fuzzy inference system, we try to analyse the previous buy and sell scenarios and try to
predict whether to go for buying or selling of stocks the next day. Since the investor
obviously wants to make profits and minimize the losses, the motivation behind this project is
to minimize the risk involved in an investment by suggesting a profitable investment plan and
keeping the investors away from a non-profitable deal based on extensive analysis of the
current market trends.

Keywords: Subtractive Clustering, Prediction, ANFIS, Stocks,

I. INTRODUCTION

Stock data analysis has remained one of the challenging time series analysis problems
over the years. Because of the complexity and instructions, the problem has received
attention from many researchers. In this paper we attempt to predict the percentage rise and
fall of the price of a specific stock index using a linear implementation of subtractive
clustering, neuro-fuzzy inference systems and a novel proposed algorithm to calculate the
contribution factor of the different clusters formed. The main strength of neuro-fuzzy

23

International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-

inference systems is that they can combine the human-like reasoning style of fuzzy systems
with the connectionist structure and learning ability of neural networks. We use subtractive
clustering because it is an easy and efficient method for automatically generating fuzzy
inference systems by detecting clusters in input-output training data. The ANFIS helps to
predict the rise and fall using previous set patterns. Using the proposed method, one can
predict the percentage rise and fall of a specific stock index. This method will be very useful
for intra-day traders as they are more interested at making small profits on a day to day basis.

II. PREVIOUS WORK

Vaidehi .V , et. al[1] build a prediction system to predict the future occurrence of an
event. It is a combination of a clustering algorithm and fuzzy system identification which
proves effective in improving the efficiency of the prediction. To train the prediction system,
historical data is obtained from the web. Data specific to any company stock is obtained and
is recorded. This recorded information is studied and parameters containing only the
necessary inputs to the prediction system. The subtractive clustering algorithm is used for its
computational advantages and fuzzy rules are formed using system identification technique.
[5] Analyses stock market price prediction based on a Generic Self-Evolving Takagi-
Sugeno-Kang (GSETSK) fuzzy neural network. Stock price prediction is a problem that
requires online adaptive systems with high accuracy performance. The proposed GSETSK
framework uses a novel Multidimensional-Scaling Growing Clustering (MSGC) algorithm
which mimics the human cognitive process to flexibly generate fuzzy rules without any a
prior knowledge.
AdemolaOlayemi (2007) mentioned that forecast performance improves when pre-
processed data is used. He has used a general pre-processing method, based on multi scale
wavelet decomposition to provide a local representation of time series data prior to the
application of fuzzy models [2].
[7] presents an innovative approach for indicating stock market decisions that the
investor should take for minimizing the risk involved in making investments. The system
uses Adaptive Neuro-Fuzzy Inference System (ANFIS) for taking decisions based on the
values of technical indicators. Among the various technical indicators available, the system
uses weighted moving averages, divergence and RSI (Relative Strength Index).
[3] reviews four of the most representative off-line clustering techniques: K-means
clustering, Fuzzy Cmeans clustering, Mountain clustering, and Subtractive clustering.

III. SUBTRACTIVE CLUSTERING

Finding similarities in data and putting similar data into groups can be done using data
clustering. Clustering partitions a data set into several groups such that the similarity within a
group is larger than that among groups [8].
Subtractive clustering is a technique for automatically generating fuzzy inference
systems by detecting clusters in input-output training data. The measure of potential for a
data point is estimated based on the distance of this data point from all other data points.
Therefore, a data point lying in a heap of other data points will have a high chance of being a
cluster centre, while a data point which is located in an area of diffused and not concentrated
data points will have a low chance of being a cluster centre.

24


After measuring the potential of every data point, the data point with the greatest
potential value is selected as the first cluster centre. To find the next cluster centre, potentials
of data points must be revised. For each data point, an amount proportional to its distance to
the first cluster centre will be subtracted. This reduces the chance of a data point near the first
cluster being selected as the next cluster centre. After revising the potential of all data points,
the data point with the maximum potential will be selected as the next cluster centre. The
potential of data points in the first step is measured as [9]:

2
n − α || x i − x j ||
pi = ∑ e − − > Equation 1
j =1

Where,
4
α =
r2

And
xi is the ithdata point and ra is a vector which consists of positive constants and
r
represents the hyper sphere cluster radius in data space. The constant a is effectively the
radius defining a neighbourhood; data points outside this radius have little influence on the
potential.
The potential which has been calculated through Equation 1 for a given point, is a
function of that point's distance to all other points, and the data point which corresponds to
*
maximum potential value is the first cluster centre. Let p1 denotes the maximum potential, if
x1* denotes the first cluster centre corresponding to p1 .
*

n
p1 = U p i − − > Equation 2
*

i =1

Where U denotes the maximum of al1
pi ' s

To revise the potential values and select the next cluster, the following formula is suggested.

* 2
* − β || xi − x j ||
pi = pi − p e 1 - - > Equation 3

Where,

4
β= 2
rb

25


And rb is a vector which consists of positive constants and is called the hyper sphere
r
penalty radius. The constant b is effectively the radius defining the neighborhood which
will have measurable reductions in potential. To avoid cluster centres being close to each
r r
other, b must be greater than a . A desirable relation is as follows [9]:

rb = 1 . 5 ra − − > Equation 4

IV. OUR CONTRIBUTION

One of the problems with subtractive clustering is that it can perform clustering
with only two factors at a time. So, in this paper we propose a simple function/algorithm
to compute the effects and contribution more than two factors. Let us call this function as
“CF Algorithm” (Contribution Factor Algorithm). The algorithm is as follows:

Combine the outcomes from both graphs.
• Construct graph 1 and graph 2 as Closing price Vs Volume and High price Vs
Low price respectively.
Initialize cluster radius CL1 and CL2 from graphs 1 and 2 respectively and the
corresponding distance of current data point as d 1 and d 2 respectively.

26


Calculate initial contribution factor R0 and new contribution factor Rn as follows:

If CL1 > CL2 then,

CL1
R0 =
CL2

 d d 
R n = R0 +  1 − 2 
 CL CL 
 1 2 

Else

CL2
R0 =
CL1

 d d 
Rn = R0 +  2 − 1 
 CL CL 
 2 1 

V. PROPOSED APPROACH

The paper proposes that if the following algorithm is used in the exact sequence then
an accurate prediction of the fall and rise of the stock index can be made. The algorithm is as
follows:

Step 1: Fix initial parameters as Closing price, Volume, High Price and Low Price.

Step 2: Obtain the historical data of the above parameters for a specific period (e.g. 5 years)
of a particular stock index.

Step 3: Map data points as Closing Price vs. Volume and High Price vs. Low Price on
different graphs.

Step 4: Perform subtractive clustering on individual graphs to get clusters and cluster centers.

Step 5: Convert the cluster into initial rules.

Step 6: Predict outcomes from individual graphs using ANFIS Modelling.

Step 7: Implement CF Algorithm.

Step 8: Stop

27


28


VI. EXAMPLE

Closing Price Vs Volume
10000000
90000000
80000000
70000000
60000000
50000000
Closing Price Vs Volume
40000000
30000000
20000000
10000000
0
0 20 40 60 80

High Vs Low
80

70

60

50

40
High Vs Low
30

20

10

0
0 10 20 30 40 50 60 70 80

Above are the graph clusters formed by 4 years of Accenture’s stock data.

29


VII. CONCLUSION/FUTURE SCOPE

The proposed paper aims to help intraday traders efficiently deal with the rise and fall
of specific stock index. Using the given algorithm, additional number of factors or parameters
can be added to further accurately predict progress of stocks. The algorithm used can be
improvised and improved for advance research.

VIII. REFERENCES

[1] Vaidehi .V, Monica .S, Mohamed Sheik Safeer .S, Deepika .M4, Sangeetha .S, “A
Prediction System Based on Fuzzy Logic”.
[2] NassimHomayouni and Ali Amiri, “Stock price prediction using a fusion model of
wavelet, fuzzy logic and ANN”.
[3] KhaledHammouda, “A Comparative Study of Data Clustering Techniques”.
[4] WeiYang, “Stock Price Prediction based on Fuzzy Logic”.
[5] Ngoc Nam Nguyen and Chai Quek, Member, IEEE, “Stock Price Prediction using
Generic Self-Evolving Takagi-Sugeno-Kang (GSETSK) Fuzzy Neural Network”.
[6] Akbar Esfahanipour, ParvinMardani, “An ANFIS Model for Stock Price Prediction: The
Case of Tehran Stock Exchange”.
[7] Samarth Agrawal, Manoj Jindal, G. N. Pillai, “Momentum Analysis based Stock Market
Prediction using Adaptive Neuro-Fuzzy Inference System (ANFIS)”
[8] Jang, J.-S. R., Sun, C.-T., Mizutani, E., “Neuro-Fuzzy and Soft Computing – A
Computational Approach to Learning and Machine Intelligence,” Prentice Hall.
[9] Chiu, S. L.; 1994, "Fuzzy model identification based on cluster estimation", Journal of
Intelligent and Fuzzy Systems, 2, John Wiley & Sons, pp. 267-278.
[10] K. V. Sujatha and S. Meenakshi Sundaram, “Regression, Theil’s and MLP
Forecasting Models of Stock Index” International journal of Computer Engineering &
Technology (IJCET), Volume 1, Issue 1, 2010, pp. 82 - 91, ISSN Print: 0976 – 6367,
ISSN Online: 0976 – 6375, Published by IAEME.

30

Novel approach for predicting the rise and fall of stock index for a specific company 2

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to Novel approach for predicting the rise and fall of stock index for a specific company 2 (20)

More from IAEME Publication (20)

Novel approach for predicting the rise and fall of stock index for a specific company 2