See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/281865047

Predicting Stock Prices Using Data Mining Techniques

Conference Paper · December 2013

CITATIONS READS

21 12,140

3 authors, including:

Qasem A. Al-Radaideh Eman Alnagi

Yarmouk University Philadelphia University Jordan
46 PUBLICATIONS   567 CITATIONS    6 PUBLICATIONS   140 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Reduct Computation View project

Arabic Natural Language Processing View project

All content following this page was uploaded by Qasem A. Al-Radaideh on 18 September 2015.

The user has requested enhancement of the downloaded file.

The International Arab Conference on Information Technology (ACIT’2013)

PREDICTING STOCK PRICES USING DATA MINING TECHNIQUES

1 2 3
QASEM A. AL-RADAIDEH, ADEL ABU ASSAF EMAN ALNAGI
1
Department of Computer Information Systems, Faculty of Information Technology and Computer Science
Yarmouk University, Irbid, Jordan. {[email protected]}
2
ICT Department, Amman Stock Exchange, Amman, Jordan. {[email protected]}
3
Department of Computer Science, Faculty of Information Technology
Philadelphia University, Jordan {[email protected]}

ABSTRACT

Forecasting stock return is an important financial subject that has attracted researchers’ attention for many years.
It involves an assumption that fundamental information publicly available in the past has some predictive
relationships to the future stock returns. This study tries to help the investors in the stock market to decide the
better timing for buying or selling stocks based on the knowledge extracted from the historical prices of such
stocks. The decision taken will be based on decision tree classifier which is one of the data mining techniques. To
build the proposed model, the CRISP-DM methodology is used over real historical data of three major companies
listed in Amman Stock Exchange (ASE).

Keywords: Data Mining, Data Mining, Data Classification, Decision Tree, Future stock return, data mining techniques,
decision tree classifiers, CRISP-DM methodology, Amman Stock Exchange.

1. INTRODUCTION various methods that aim to predict future price

movements using past stock prices and volume
The stock market is essentially a non-linear, non- information. It is based on the assumption that history
parametric system that is extremely hard to model with repeats itself and that future market directions can be
any reasonable accuracy [1]. Investors have been trying to determined by examining historical price data. Thus, it is
find a way to predict stock prices and to find the right assumed that price trends and patterns exist that can be
stocks and right timing to buy or sell. To achieve those identified and utilized for profit. Most of the techniques
objectives, and according to [2], [3-4] some research used used in technical analysis are highly subjective in nature
the techniques of fundamental analysis, where trading and have been shown not to be statistically valid.
rules are developed based on the information associated
with macroeconomics, industry, and company. The Recently, data mining techniques and artificial
authors of [5] and [6] said that fundamental analysis intelligence techniques like decision trees, rough set
assumes that the price of a stock depends on its intrinsic approach, and artificial neural networks have been applied
value and expected return on investment. Analyzing the to this area [8]. Data mining [9] refers to extracting or
company’s operations and the market in which the mining knowledge from large data stores or sets. Some of
company is operating can do this. Consequently, the stock its functionalities are the discovery of concept or class
price can be predicted reasonably well. Most people descriptions, associations and correlations, classification,
believe that fundamental analysis is a good method only prediction, clustering, trend analysis, outlier and deviation
on a long-term basis. However, for short- and medium- analysis, and similarity analysis. Data classification can
term speculations, fundamental analysis is generally not be done in many different methods; one of those methods
suitable. is the classification by using Decision Tree. It is a
graphical representation of all possible outcomes and the
Some other research used the techniques of technical paths by which they may be reached.
analysis [2], in which trading rules were developed based
on the historical data of stock trading price and volume. Decision trees and artificial neural networks can be
Technical analysis as illustrated in [5] and [7] refers to the trained by using an appropriate learning algorithm.
Following the assumption of technical analysis that data, data mining techniques are new techniques that can
patterns exist in price data, it is possible in principle to be used to extract the knowledge from this data.
use data mining techniques to discover these patterns in
an automated manner. Once these patterns have been For that reason, several researchers have focused on
discovered, future prices can be predicted. technical analysis and using advanced math and science.
Extensive attention has been dedicated to the field of
Today, the grand challenge of using a database is to artificial intelligence and data mining techniques [11].
generate useful rules from raw data in a database for users Some models have been proposed and implemented using
to make decisions, and these rules may be hidden deeply the above mentioned techniques, the authors of [5] made
in the raw data of the database. Traditionally, the method an empirical study on building a stock buying/selling alert
of turning data into knowledge relies on manual analysis; system using back propagation neural networks (BPNN),
this is becoming impractical in many domains as data their NN was codenamed NN5. The system was trained
volumes grow exponentially. The problem with predicting and tested with past price data from Hong Kong and
stock prices is that the volume of data is too large and Shanghai Banking Corporation Holdings over the period
huge. This paper uses one of the data mining methods; from January 2004 to December 2005. The empirical
which is the classification approach on the historical data results showed that the implemented system was able to
available to try to help the investors to build their decision predict short-term price movement directions with
on whether to buy or sell that stock in order to achieve accuracy about 74%.
profit.
The research by [2] used decision tree technique to build
The main objective of this paper is to analyze the on the work of Lin [12] where Lin tried to modify the
historical data available on stocks using decision tree filter rule that is to buy when the stock price rises k%
technique as one of the classification methods of data above its past local low and sell when it falls k% from its
mining in order to help investors to know when to buy past local high. The proposed modification to the filter
new stocks or to sell their stocks. rule in [12] was by combining three decision variables
associated with fundamental analysis. An empirical test,
Analyzing stock price data over several years may involve using the stocks of electronics companies in Taiwan,
a few hundreds or thousands of records, but these must be showed Lin’s method outperformed the filter rule.
selected from millions. The data that will be used in this According to [2], in Lin’s work, the criteria for clustering
paper to build the decision tree will be the historical trading points involved only the past information; the
prices of three listed companies in Amman Stock future information was not considered at all. The research
Exchange over two years of time. by [2] aimed to improve the filter rule and Lin’s study by
considering both the past and the future information in
The remainder of this paper is organized into four clustering the trading points. The researchers used the
sections. Section 2 of the paper gives a literature review data of Taiwan stock market and that of NASDAQ to
about the subject of using data mining techniques in order carry out empirical tests. Test results showed that the
to try to predict the prices and the trend of stocks, some proposed method outperformed both Lin’s method and the
related work in that subject is shown in this section. filter rule in the two stock markets.
Section 3 talks about the methodology used in building
the classification model. Then section 4 shows the The model of [11] applied the concept of serial topology
experiments that are done on the data collected using the and designed a new decision system, namely the two-
model and evaluation of the results using one of the layer bias decision tree, for stock price prediction. The
evaluation methods. Finally, a brief conclusion and the methodology developed by the authors differs from other
future work about the topic is given in section 5. studies in two respects; first, to reduce the classification
error, the decision model was modified into a bias
2. LITERATURE REVIEW decision model. Second, a two-layer bias decision tree is
used to improve purchasing accuracy. The empirical
Over the past two decades many important changes have results indicated that the presented decision model
taken place in the environment of financial markets. The produced excellent purchasing accuracy, and it
development of powerful communication and trading significantly outperformed than random purchase.
facilities has enlarged the scope of selection for investors.
Forecasting stock return is an important financial subject The authors of [10] presented an approach that used data
that has attracted researchers’ attention for many years. It mining methods and neural networks for forecasting stock
involves an assumption that fundamental information market returns. An attempt has been made in this study to
publicly available in the past has some predictive investigate the predictive power of financial and
relationships to the future stock returns [10]. In order to economic variables by adopting the variable relevance
be able to extract such relationships from the available analysis technique in machine learning for data mining.

2
The authors examined the effectiveness of the neural simulate results to confirm the theoretical computations of
network models used for level estimation and the approach.
classification. The results showed that the trading
strategies guided by the neural network classification 3. THE METHODOLOGY OF THE STUDY
models generate higher profits under the same risk
exposure than those suggested by other strategies. Data mining methodology is designed to ensure that the
data mining effort leads to a stable model that
The research by [13] was basically a comparison between successfully addresses the problem it is designed to solve.
the work of Fama and French’s model [14-15] and the Various data mining methodologies have been proposed
artificial neural networks in order to try to predict the to serve as blueprints for how to organize the process of
stock prices in the Chinese market. The purpose of this gathering data, analyzing data, disseminating results,
study is to demonstrate the accuracy of ANN in predicting implementing results, and monitoring improvements [9].
stock price movement for firms traded on the Shanghai To build the model that analyses the stock trends using
Stock Exchange. In order to demonstrate the accuracy of the decision tree technique, the CRISP-DM (Cross-
ANN, the authors made a comparative analysis between Industry Standard Process for data mining) [20] is used.
Fama and French’s model and the predictive power of the This methodology was proposed in the mid-1990s by an
univariate and multivariate neural network models. The European consortium of companies to serve as a non-
results from this study indicated that artificial neural proprietary standard process model for data mining. This
networks offer an opportunity for investors to improve model consists of the following six steps:
their predictive power in selecting stocks, and more
importantly, a simple univariate model appears to be more
 Understanding the reason and objective of mining the
successful at predicting returns than a multivariate model.
stock prices.
 Understanding the collected data and how it is
Al-Haddad et al., [16] presented a study that aimed to
structured.
provide evidence of whether or not the corporate
governance & performance indicators of the Jordanian  Preparing the data that is used in the classification
industrial companies listed at Amman Stock Exchange model.
(ASE) are affected by variables that were proposed and to  Selecting the technique to build the model.
provide the important indicators of the relationship of  Evaluating the model by using one of the well known
corporate governance & firms’ performance that can be evaluation methods.
used by the Jordanian industrial firms to solve the agency  Deploying the model in the stock market to predict
problem. The study random sample consists of (44) the best action to be taken, either selling or buying
Jordanian industrial firms. The study founds a positive the stocks.
direct relationship between corporate governance and  Understanding the reason and objective of building
corporate performance. the model

Hajizadeh et al. [17] provided an overview of application The main reason and objective of building the model is to
of data mining techniques such as decision tree, neural try to help the investors in the stock market to decide the
network, association rules, and factor analysis and in best timing for buying or selling stocks based on the
stock markets. knowledge extracted from the historical prices of such
stocks. The decision taken will be based on one of the
Prediction stock price or financial markets has been one data mining techniques; the decision tree classifiers.
of the biggest challenges to the AI community. Various
technical, fundamental, and statistical indicators have Understanding the collected data
been proposed and used with varying results. Soni [18]
surveyed some recent literature in the domain of machine The Oracle database of Amman Stock Exchange (ASE)
learning techniques and artificial intelligence used to contains the historical prices of the 230 companies listed
predict stock market movements. Artificial Neural in the exchange from the year 2000. As the amount of
Networks (ANNs) are identified to be the dominant such data is very large and complicated, the decision was
machine learning technique in stock market prediction taken to choose three companies listed in the exchange.
area. The selection of these companies was based on the
following five criteria which represent the companies' size
El-Baky et al., [19], proposed a new approach for fast and liquidity: Market capitalization, days traded, turnover
forecasting of stock market prices. The proposed ratio, value traded and the number of shares traded, also
approach uses new high speed time delay neural networks the sector representation was considered during the
(HSTDNNs). The authors used the MATLAB tool to selection of these companies. These companies are “Arab
Bank”, its’ code in the stock market “ARBK” and it

3
belongs to the banking sector, “United Arab Investors other attributes were found not important and not having a
Company”, its’ code is “UAIC” and it belongs to the direct effect on the study. Table1 shows the 6 attributes
services sector, and “Middle East Complex for selected with their descriptions and their possible values.
Engineering, Electronics and Heavy Industries”, its’ code The class attribute is the investor action whether to buy or
is “MECE” and it belongs to the industrial sector. The sell that stock and it is named, “Action”. The data of this
period that was selected is from April 2005 to May 2007, attribute was taken also from ASE database, which is the
which presented the current and actual status of the net position of one of the biggest brokers dealing with the
market at that period of time. above mentioned stocks every day. The net position could
be either buying or selling that stock for that day.
At the beginning, the data collected contained 9 attributes;
this number was reduced manually to 6 attributes as the
Table 1: Attribute Description

Attribute Description Possible Values

Previous Previous day close price of the stock Positive, Negative, Equal
Open Current day open price of the stock Positive, Negative, Equal
Min Current day minimum price of the stock Positive, Negative, Equal
Max Current day maximum price of the stock Positive, Negative, Equal
Last Current day close price of the stock Positive, Negative, Equal
Action The action taken by the investor on this stock Buy, Sell

Preparing the data generalizing them to discrete values, while table3 shows
the same sample after selecting the 6 attributes and after
At the beginning, when the data was collected, all the transforming them to discrete values.
values of the attributes selected were continuous numeric
values. Data transformation was applied by generalizing Building the model
data to a higher-level concept so as all the values became
discrete. The criterion that was made to transform the After the data has been prepared and transformed, the
numeric values of each attribute to discrete values next step was to build the classification model using the
depended on the previous day closing price of the stock. If decision tree technique. The decision tree technique was
the values of the attributes open, min, max, last were selected because [9] the construction of decision tree
greater than the value of attribute previous for the same classifiers does not require any domain knowledge, thus it
trading day, the numeric values of the attributes were is appropriate for exploratory knowledge discovery. Also,
replaced by the value Positive. If the values of the it can handle high dimensional data. Another benefit is
attributes mentioned above were less than the value of the that the steps of decision tree induction are simple and
attribute previous, the numeric values of the attributes fast. Generally, decision tree accuracy is considered good.
were replaced by Negative. If the values of those The decision tree method depends on using the
attributes were equal to the value of the attribute previous, information gain metric that determines the most useful
the values were replaced by the value Equal. Table 2 attribute. The information gain depends on the entropy
shows a sample of the continuous numeric values of the measure.
data before selecting the 6 attributes manually and before

Table 2: Sample of historical data before selecting relevant attributes and before generalization

Previous Open Max Min Last Action

25.82 25.99 26 25.41 25.67 Sell
25.67 25.68 25.68 25.2 25.3 Buy
25.3 24.8 25.3 24.41 24.9 Buy
24.9 24.8 24.9 24.3 24.87 Sell
24.87 24.87 25.55 24.85 25.3 Buy
25.3 25.25 26 25.25 25.82 Buy
25.82 25.99 26.4 25.99 26.3 Buy

4
 26.3 26.3 26.7 26 26.02 Buy
26.02 26.09 26.25 25.55 25.63 Sell
Table 3: Sample of historical data after selecting attributes and after generalization.

Previous Open Max Min Last Action

Positive Positive Positive Negative Negative Sell
Negative Positive Positive Negative Negative Buy
Negative Negative equal Negative Negative Buy
Negative Negative equal Negative Negative Sell
Negative equal Positive Negative Positive Buy
Positive Negative Positive Negative Positive Buy
Positive Positive Positive Positive Positive Buy
Positive equal Positive Negative Negative Buy
Negative Positive Positive Negative Negative Sell

The gain ratio is used to rank attributes and to build the classification rules that resulted after building the decision
decision tree where each attribute is located according to trees for each company using the C4.5 algorithm.
its gain ratio. When the decision tree model was applied
on the data of the three companies using the WEKA The graphs of the resulting decision trees using the C4.5
software version 3.5 [21], the root attribute for both algorithm with pruning technique is presented in Figure 1,
ARBK and UAIC company was the Open, while the Figure 2, and Figure 3 for the three companies under
attribute Last was the root for the decision tree of the study.
MECE company. As the process of building the tree goes
Table 4: Summary of the number of the classification rules
on, all the remaining attributes were used to continue with
this process. After building the complete decision tree, the Number of
Number of classification
set of classification rules were generated by following all Company classification rules
rules with pruning
without pruning
the paths of the tree. The maximum number of attributes
ARBK 21 11
that were used in some of the classification rules
generated were 4 attributes, while some classification UAIC 31 5
rules used only 1 attribute. Both the ID3 and C4.5
algorithms were used in building the decision trees and MECE 21 9
the pruning technique was used in the C4.5 algorithm in
order to reduce the size of the produced decision trees.
Table 4 gives a summary about the numbers of the

Figure 1: Decision Tree for the MECE

5
 Figure 2: The Decision Tree for the ARBK

4. RESULTS AND DISCUSSION

This section presents step 5 of the CRISP methodology
which was used to build the model. It is simply about
evaluating the model by using one or more of the well
known evaluation methods. In order to evaluate the
model, the WEKA software was used to calculate the
accuracy of the classification model. Two evaluation
methods were used, the K-Fold Cross Validation (K-CV)
where K=10 folds and the percentage split method where
66% of the data was used for training and the remainder
for testing. Both evaluation methods were used on the ID3
and C4.5 decision tree classification methods. Table 5
shows the accuracy of all the classifiers generated using
both classification methods and both evaluation methods.

As we can see from the table, the resultant classification

Figure 3: The Decision Tree for the UAIC accuracy from the decision tree model is not very high for
the training data used and it varies from one company to
another. The reason for such a low accuracy is that the
Deploying the model company’s performance in the stock market is affected by
internal financial factors such as; news about the
The classification rules that were generated from the company, financial reports, and the overall performance
decision tree model can be used and integrated in a of the market. Also, external factors can affect the
system that predict the best action and timing for the performance of the company in the market such as;
investors, either to buy or sell the stocks on that day. political events and political decisions. Thus, it can be
difficult to have a model that gives a high accuracy
classification for all the companies at the same time as the
performance of these companies differs.

6
 Table 5: Classification accuracy using ID3 & C4.5 classification methods and using 10-CV& Holdout evaluation methods

10-CV Holdout 66%

Classification Total Correctly Accuracy Total Correctly Accuracy

Company Instances Instances
Method classified % classified %
ID3 233 44.689 73 42.941
ARBK 499 170
C4.5 237 47.495 83 48.824
ID3 255 50.797 84 49.123
MECE 502 171
C4.5 265 52.789 91 53.216
ID3 269 53.586 88 51.462
UAIC 502 171
C4.5 264 52.590 94 54.971

5. CONCLUSIONS AND FUTURE WORK [5] Tsang, P.M., Kwok, P., Choy, S.O., Kwan, R., Ng,
S.C., Mak, J., Tsang, J., Koong, K., and Wong, T.L.
This study presents a proposal to use the decision tree (2007) “Design and implementation of NN5 for Hong
classifier on the historical prices of the stocks to create Kong stock price forecasting”, Engineering Applications
decision rules that give buy or sell recommendations in of Artificial Intelligence, 20, pp. 453-461.
the stock market. Such proposed model can be a helpful [6] Ritchie, J.C., (1996) Fundamental Analysis: a Back-
tool for the investors to take the right decision regarding to-the-Basics Investment Guide to Selecting Quality
their stocks based on the analysis of the historical prices Stocks. Irwin Professional Publishing.
of stocks in order to extract any predictive information
from that historical data. The results for the proposed [7] Murphy, J.J., (1999) Technical Analysis of the
model were not perfect because many factors including Financial Markets: a Comprehensive Guide to Trading
but not limited to political events, general economic Methods and Applications. New York Institute of
conditions, and investors’ expectations influence stock Finance.
market. [8] Wang, Y.F., (2002) “Predicting stock price using
fuzzy grey prediction system”, Expert Systems with
As for the future work, there is still big room for testing Applications, 22, pp. 33-39.
and improving the proposed model by evaluating the
model over the whole companies listed in the stock [9] Han, J., Kamber, M., Jian P. (2011). “Data Mining
market. Also, the evaluation of a larger collection of Concepts and Techniques”. San Francisco, CA: Morgan
learning techniques such as neural networks, genetic Kaufmann Publishers.
algorithms, and association rules can represent a rich area [10] Enke, D., Thawornwong, S. (2005) “The use of data
for future investigation. Finally, reconsidering the factors mining and neural networks for forecasting stock market
affecting the behavior of the stock markets, such as returns”, Expert Systems with Applications, 29, pp. 927-
trading volume, news and financial reports which might 940.
impact stock price can be another rich field for future
studying. [11] Wang, J.L., Chan, S.H. (2006) “Stock market trading
rule discovery using two-layer bias decision tree”, Expert
REFERENCES Systems with Applications, 30(4), pp. 605-611.
[12] Lin, C. H. (2004) Profitability of a filter trading rule
[1] Wang, Y.F., (2003) “Mining stock price using fuzzy on the Taiwan stock exchange market. Master thesis,
rough set system”, Expert Systems with Applications, 24, Department of Industrial Engineering and Management,
pp. 13-23. National Chiao Tung University.
[2] Wu, M.C., Lin, S.Y., and Lin, C.H., (2006) “An [13] Cao, Q., Leggio, K.B., and Schniederjans, M.J.,
effective application of decision tree to stock trading”, (2005) “A comparison between Fama and French’s model
Expert Systems with Applications, 31, pp. 270-274. and artificial neural networks in predicting the Chinese
[3] Al-Debie, M., Walker, M. (1999). “Fundamental stock market”, Computers & Operations Research, 32, pp.
information analysis: An extension and UK evidence”, 2499-2512.
Journal of Accounting Research, 31(3), pp. 261–280. [14] Fama, E.F., French, K.R., (1993) “Common risk
[4] Lev, B., Thiagarajan, R. (1993). “Fundamental factors in the returns on stocks and bonds”, The Journal of
information analysis”, Journal of Accounting Research, Finance, 33, pp. 3-56.
31(2), 190–215.

7
 [15] Fama, E.F., French, K.R., (1992) “The cross-section [19] Hazem M. El-Bakry, and Wael A. Awad, Fast
of expected stock returns”, The Journal of Finance, 47, Forecasting of Stock Market Prices by using New High
pp. 427-465. Speed Time Delay Neural Networks, International Journal
of Computer and Information Engineering 4:2 2010. Pp
[16] Al-Haddad W. Alzurqan S. and Al_Sufy S, The
138-144.
Effect of Corporate Governance on the Performance of
Jordanian Industrial Companies: An empirical study on [20] Chapman P., Clinton J., Kerber R., Khabaza T.,
Amman Stock Exchange. International Journal of Reinartz T., Shearer C., and Wirth R., (2000). “CRISP-
Humanities and Social Science, Vol. 1 No. 4; April 2011 DM 1.0: Step-by-step data mining guide”.
[17] Hajizadeh E., Ardakani H., and Shahrabi J., [21] Witten I. Frank E., and Hall M. (2011), “Data
Application of data mining techniques in stock markets: A Mining: Practical Machine Learning Tools and
survey, Journal of Economics and International Finance Techniques”, 3rd Edition, Morgan Kaufmann Publishers.
Vol. 2(7), pp. 109-118, July 2010.
[18] Soni S., Applications of ANNs in Stock Market
Prediction: A Survey, International Journal of Computer
Science & Engineering Technology (IJCSET), pp 71-83,
Vol. 2 No. 3, 2011.

View publication stats