T0 numtq0n tk=

International Journal of Science and Research (IJSR)
ISSN (Online): 2319-7064
Impact Factor (2012): 3.358
An Approach to Analyze Stock Market Using
Clustering Technique
R. Suganthi1. Dr. P. Kamalakannan2
1M.C.A, M.Phil, Department of Computer Applications, Valluvar College of Science & Management, Karur, India
2M.C.A, Ph.D., Department of Computer Applications, Government Arts College, Salem, India
Abstract: In order to analyze the various sectors of the stocks in stock market field, need to use machine learning algorithms to
determine the particular sectors of the stocks in intraday trade. In this paper, we compared different types of clustering algorithm with
the help of data mining tool WEKA. This paper will demonstrate the strength and accuracy of each algorithm for clustering in terms of
performance, efficiency and time complexity required
Keyword: Machine Learning, Clustering, Weka tool, Multi database, Stock market data
1. Introduction
Machine learning has confirmed to be of good quality value
in different kind of application domains. It is specifically
useful in data mining problems where big databases may
have valuable enclosed regularities that can be discovered
automatically. Designing a machine learning approach
contains no of design options like choosing the typing of
training experience. Target function from training ex:
various commonly used machine learning algorithms are
Artificial neural network, decision tree, genetic algorithm,
apriori algorithm, Rule induction etc..The application of
machine learning approach to stock market data is a recent
trend in research. The stock market daily trade result in
stock market field has been still a source of great concern
and research interest to process the analyzing stock market
data and buyers to find the better stocks in different stock
sectors. The discovery of this process can help the customer
to take their own decision on daily basis trade where it can
analyze the buying habit of the customers.
The literature is replete with various works in a machine
learning area on stock market data. The focus of this work is
on applying machine learning algorithms to stock market
data for predictive and analyzing data purposes in stock
market field.500 records of dataset has to be used the
training data will be measured by clustering algorithm. The
comparison of various algorithm will be produced the
performance and efficiency of dataset.
2. Aim & Objective
The aim of this study is to use machine learning algorithms
to determine the specific sector stocks in intraday trade of
stock market, the general objectives is to demonstrate how
machine learning algorithms can be applied to stock market
data.
 To create database representing the specific sector of stock
market
 To model the daily trade data based on dataset.
 To compare models generated from the machine learning
algorithms (K-means, optics, EM, Cobweb) using the
accuracy level and Time taken and identify which model
is most appropriate for taking decision on daily trade data.
 To generate the model.
 To examine the sectors to see what can be learned from
that
 To design a system framework based on efficiency of
algorithm.
3. Literature Review
There has been significant progress in the field of machine
learning bordering on its application to the areas of
medicine, natural language processing, software
development and inspection, financial investing, stock
market applications and so on Edwin Lughofer et al, 2012
[3] uses cluster analysis to group related documents for
browsing, to find genes and proteins that have similar
functionality, and to provide a grouping of spatial locations
prone to earthquakes. The ability to discover highly
correlated regions of objects when their number becomes
very large is highly desirable, as data sets grow and their
properties and data interrelationships change.
Kevin Duh [4] has proposed a novel active learning strategy
for data-driven classifiers, which is based on unsupervised
criterion during off-line training phase, followed by a
supervised certainty-based criterion during incremental on-line
training. In this se, they call the new strategy hybrid
active learning. Sample selection in the first phase is
conducted from scratch (i.e.no initial labels/learners are
needed) based on purely unsupervised criteria obtained from
clusters: samples lying near cluster centres and near the
borders of clusters are expected to represent the most
informative ones regarding the distribution characteristics of
the classes. In the second phase, the task is to update already
trained classifiers during on-line mode with the most
important samples in order to dynamically guide the
classifier to more predictive power.
Both strategies are essential for reducing the annotation and
supervision effort of operators in off-line and on-line
classification systems, as operators only have to label an
exquisite subset of the off-line training data representation
Volume 3 Issue 10, October 2014
www.ijsr.net
Paper ID: OCT14459 2234
Licensed Under Creative Commons Attribution CC BY

give feedback only on specific occasions during on-line
phase.
Xiaodong Yu et al. [5] have proposed a flexible transfer
learning strategy based on sample selection. Source domain
training samples are selected if the functional relationship
between features and labels do not deviate much from that of
the target domain. This is achieved through a novel
application of recent advances from density ratio estimation.
The approach is flexible, scalable, and modular. It allows
many existing supervised rankers to be adapted to the
transfer learning setting. R.J. Gil. [6] have proposed a novel
updating algorithm based on iterative learning strategy for
delayed coking unit (DCU), which contains both continuous
and discrete characteristics. Daily DCU operations under
different conditions are modelled by a belief rule-base
(BRB), which is then, updated using iterative learning
methodology, based on a novel statistical utility for every
belief rule. Compared with the other learning algorithms,
their methodology can lead to a more optimal compact final
BRB. With the help of this expert system, a feed forward
compensation strategy is introduced to eliminate the
disturbance caused by the drum-switching operations.
R.J. Gil et al. [6] have proposed a novel model of an
Ontology-Learning Knowledge Support System (OLeKSS)
is proposed to keep these KSSs updated. The proposal
applies concepts and methodologies of system modelling as
well as a wide selection of OL processes from
heterogeneous knowledge sources (ontology texts, and
databases), in order to improve KSS’s semantic product
through a process of periodic knowledge updating. An
application of a Systemic Methodology for OL (SMOL) in
an academic Case Study illustrates the enhancement of the
associated ontology through process of population and
enrichment.
Md Nasir et al. have proposed a variant of single-objective
PSO called Dynamic Neighbourhood Learning Particle
Swarm Optimizer (DNLPSO), which uses learning strategy
whereby all other particles’ historical best information is
used to update a particle’s velocity as in Comprehensive
Learning Particle Swarm Optimizer (CLPSO). But in
contrast to CLPSO, in DNLPSO, the exemplar particle is
selected from a neighbourhood. This strategy enables the
learner particle to learn from the historical information of its
neighbourhood or sometimes from that of its own.
4. Methodology
In this research, we focused on comparing the performance
of machine learning algorithms that are trained with data
relating to intraday trade with the aim of obtaining good
stocks sectors for taking decision by their own of customers
that will provide the proper alignment of daily basis trading
data. For collection data to train the machine learning
algorithms, quantitative approach will be used. A tool will
be built based on database that interfaces with weka for
training the algorithm for finding the optimization of
analyzing model of stock market data will make a
comparative research on the machine learning algorithms
using cross validation and non clustering errors benchmarks.
For training 40% date will be used while the remaining 60%
will be used to validate. During data collection, the relevant
data will be gathered. Once the data has been collected, its
quality will be verified. Incomplete data will be eliminated
and the data will be cleaned by filling in missing values,
smoothing noisy data, identifying or removing outliers and
resolving in consistencies. At last, the Noisy data will be
stored in different tables and later joined in a single table to
remove errors.
a. Various Techniques
There are several clustering algorithm available in weka. But
cobweb, DB scan, EM, Optics and K-Means will be used in
this study. Attribute importance analysis will be carried out
to rank the attributes by significance using information gain.
Consistency subset selection and feature subset selection
filter algorithm will be used to rank and select the attributes
that are mostly used. The COBWEB algorithm was
developed by machine learning researchers in the 1980s for
clustering objects in a object-attribute data set. The
COBWEB algorithm yields a clustering dendrogram called
classification tree that characterizes each cluster with a
probabilistic Description. Cobweb generates hierarchical
clustering, where clusters are described probabilistically.
COBWEB uses a heuristic evaluation measure called
category utility to guide construction of the tree. It
incrementally incorporates objects into a classification tree
in order to get the highest category utility.
b. B. Experimental Results
The data set is with COBWEB algorithm with evaluation on
training set of WEKA tools. The number of merges found
was 76, and number of splits are 71. Total time taken to
build model on full training data was 27.88 seconds.
Numbers of clusters were 107. See figure 1.
Figure 1: Result of cobweb clustering algorithm
Figure 2: Result of cobweb cluster in form of graph
www.ijsr.net

c. Db scan Clustering Algorithm
DBSCAN (for density-based spatial clustering of
applications with noise) is a data clustering Algorithm
proposed by Martin Ester, Hans-Peter Kriegel, Jorge Sander
and Xiaowei Xu in 1996. It is a density-based clustering
algorithm because it finds a number of clusters starting from
the estimated density distribution of corresponding nodes.
DBSCAN is one of the most common clustering algorithms
and also most cited in scientific literature. OPTICS can be
seen as a generalization of DBSCAN to multiple ranges,
effectively replacing the parameter with a maximum search
radius.
d. Experimental Results
The NSE data set is applied with COBWEB algorithm for
evaluation on training set with Epsilon: 0.9, min Points: 6.
Distance-type of clusters for DBSCAN Data Objects is
Euclidian Data Object. The Number of generated clusters
are 4 and elapsed time is 1.14 seconds.
e. Hierarchical Clustering
In hierarchical clustering the data are not partitioned into a
particular cluster in a single step. Instead, a series of
partitions takes place, which may run from a single cluster
containing all objects to n clusters each containing a single
object. Hierarchical Clustering is subdivided into
agglomerative methods, which proceed by series of fusions
of the n objects into groups, and divisive methods, which
separate n objects successively into finer groupings.
Hierarchical clustering may be represented by a two
dimensional diagram known as dendrogram which illustrates
the fusions or divisions made at each successive stage of
analysis.
f. Experimental Results
The NSE data set is applied with hierarchical clustering
algorithm for evaluation on training set with four clusters.
The elapsed time is 1.88 seconds. The tree Visualizer can
also be analyzed
g. The K Means Clustering
The k-means algorithm is one of a group of algorithms
called partitioning methods. The k –means algorithm is very
simple and can be easily implemented in solving many
practical problems. The k-means algorithm is the best-known
squared error-based clustering algorithm.
1. Selection of the initial k means for k clusters
2.Calculation of the dissimilarity between an object and the
mean of a cluster
3. Allocation of an object to the cluster whose mean is
nearest to the object
4. Re-calculation of the mean of a cluster from the objects
allocated to it so that the intra Cluster dissimilarity is
minimized. Except for the first operation, the other three
operations are repeatedly performed in the algorithm until
the algorithm converges. The essence of the algorithm is
to minimize the cost function where n is the number of
objects in a data set X, Xi €
X, Ql is the mean of cluster l, and Yi l is an element of a
partition matrix Yn x l as in (Hand 1981). d is a dissimilarity
measure usually defined by the squared Euclidean distance.
There exist a few variants of the k-means algorithm which
differ in selection of the initial k means, dissimilarity
calculations and strategies to calculate cluster means. The
sophisticated variants of the k-means algorithm include the
well-known ISODATA algorithm and the fuzzy k-means
algorithms.
h. Experimental Results
The NSE data set is applied with k-means algorithm for
evaluation on training set. Distance-type of Clusters for k-means
Data Objects is Euclidian Distance. The number of
iterations is 17, within cluster sum of squared errors:
127.02034887051619. Missing values globally replaced
with mean/mode The Number of generated clusters are 4
and elapsed time is 0.16 seconds. See fig 7.
i. Comparisons of various algorithms
Cobweb:
No of instances: 420
No of attributes: 6
No of merges: 76
No of splits: 71
No of clusters: 107
EM
No of clusters elected by cross validation:4
Cluster Instances: 4
Log likelihood: 14.01445
DBSCAN
Clustering data objects: 420
No of generated clusters: 21
Elapsed time: 3.97
Un clustered instances: 40
K-Means:
No of iterations: 4
Within cluster sum of squared errors: 539.34
Clustered instances: 0 372 89%
1 48 11%
5. Conclusion
The main aim of this paper is to provide a detailed
introduction of weka clustering algorithms.. It is the simplest
tool for classify the data of various types. It is the first model
to provide the graphical user interface of the user in order to
performing the clusterization with the help of NSE data. It
provides some features to analyze the NSE dataset for
intraday trading with different clustering algorithms.
Comparatively k-Means is the best because of simplicity and
fastest than other algorithms that why k-means clustering
algorithm is more suitable for stock market data mining
applications. This paper shows only the clustering operations
in the weka, further research can be performed using other
data mining algorithms on Stock market data. In future, we
can use various data mining algorithms like classification,
association to produce good result by more records.
References
[1] Edwin Lughofer. Hybrid active learning for reducing the
annotation effort of operators in classification systems,
Pattern Recognition 45 (2012) 884–896.
www.ijsr.net

[2] Han J. and Kamber M.: “Data Mining: Concepts and
Techniques,” Morgan Kaufmann Publishers, San
Francisco,2000.
[3] Kevin Duh, Akinori Fujino. Flexible sample selection
strategies for transfer learning in ranking, Information
Processing and Management 48 (2012) 502–512.
[4]Md Nasir, Swagatam Das, Dipankar Maity, Soumyadip
Sengupta, Udit Halder, P.N. Suganthan. A dynamic
neighborhood learning based particle swarm optimizer
for global numerical optimization, Information Sciences
209 (2012) 16–36.
[5] R.J. Gil, M.J. Martin-Bautista. A novel integrated
knowledge support system based on ontology learning:
Model specification and a case study, Knowledge-Based
Systems 36 (2012) 340–352.
[6] Sapna Jain, M Afshar Aalam and M N Doja, “K-means
clustering using weka interface”, Proceedings of the 4th
National Conference; INDIACom-2010
[7]Xiaodong Yu , DexianHuang, YonghengJiang, YihuiJin.
Iterative learning belief rule-base inference methodology
using evidential reasoning for delayed coking unit,
Control Engineering Practice 20 (2012) 1005–1015.
www.ijsr.net

T0 numtq0n tk=

More Related Content

What's hot (20)

Viewers also liked (19)

Similar to T0 numtq0n tk= (20)

More from International Journal of Science and Research (IJSR) (20)

T0 numtq0n tk=