Download Static Data Mining Algorithm with Progressive

Static Data Mining Algorithm with Progressive Approach for
Mining Knowledge
Shilpa#1 and Sunita Parashar *2
#
Student, Department of Computer Science & Engineering
Haryana College of Technology & Management, Kaithal, Haryana, India
E-mail: [email protected]
*
Associate Professor, Department of Information Technology,
Haryana College of Technology & Management, Kaithal, Haryana, India
E-mail: [email protected]
Abstract
Frequent itemsets generation is an important area of data
mining. This paper is concerned with applying progressive
approach to extract interesting information from a static
database using dynamic approach. This provides an intelligent
environment to discover frequent itemsets while reading a
particular set of transaction from static database. We
performed extensive experiments and calculate the execution
time to generate frequent itemsets on the basis of support and
number of transaction read at a time.
Keywords: Static data mining, Dynamic data mining,
Support, Number of transactions read at a time, Execution
time.
Introduction
With the rapid growth in size and number of available
databases in commercial, industrial, administrative and other
applications, it is necessary and interesting to examine how to
extract knowledge automatically from huge amount of data
[1]. Knowledge discovery in databases (KDD), or Data
Mining, is the effort to understand, analyze, and eventually
make use of huge volume of data available. Data mining is the
discovery of hidden information found in databases and can be
viewed as a step in overall process of Knowledge Discovery in
databases (KDD) [2][3]. It is the integration of various
techniques from multiple disciplines such as statistics,
machine learning, pattern recognition, neural networks, image
processing, and database management system and so on[4]. It
makes use of various algorithms to perform a variety of tasks.
These algorithms examine the sample data of a problem and
determine a model that fits close to solving the problem. The
models that we determine to solve a problem are classified as
predictive and descriptive [5][6]. Predictive mining tasks
perform inference on current data in order to make
predictions. The data mining task that forms the part of
predictive model are Classification, Regression, and Time
series analysis. Descriptive mining tasks characterize the
general properties of the data in the database. This enables us
to determine the patterns and relationships in a sample data. A
data mining task that forms the part of descriptive model are
Clustering, Summarization, Association rules, Sequence
discovery. Classification derives a function or model that
describes and distinguishes data classes or concepts, which
determines the class of objects whose class label is unknown.
The derived model is based on the analysis of a set of training
data. The training data includes data objects whose class label
is known. Regression is to forecast future data values based on
present and past data values by means of mathematical
formula. Time series analysis is to predict future values for
current set of values that are time dependent. Clustering
identifies the classes also called clusters or groups for the set
of objects whose classes are unknown. The objects are so
clustered that the intraclass similarities are maximized and the
interclass similarities are minimized. This is done based on the
criteria defined on the attributes of the objects [1][6].
Summarization is the abstraction or generalization of data.
This results in a smaller set, which gives a general overview of
data, usually with aggregated information. Summarization is
used to summarize huge amount of data containing in a web
page or document. The summarization can go to different
abstraction levels and can be viewed from different angles. It
is also known as characterization or generalization.
Association rule mining is to generate correlation between
large unclassified data items based on certain attributes and
characteristics and association rule. Association rules are used
to identify relationship among a set of items in database of
transactions on the basis of large itemsets [7]. Sequence
discovery is to determine the sequential patterns that exist in
data by using the time factor.
In data mining, with the increasing amount of data stored
in real application system, the discovery of association
relationship (Association Rule mining) attracts more and more
attention. Mining for association rules can help in business,
decision making, and the development of customized
marketing programs and strategies. Thus goal of data mining
is to turn “data into knowledge” [8].Therefore, mining
association rules from large database has been a focused topic
in recent research into knowledge discovery in databases [9].
Database can be static and dynamic. Static databases are
those databases that do not change with time while in dynamic
databases, new transactions append as time advances. This
may introduce new frequent itemsets and some existing
[Page No. 464]
Static Data Mining Algorithm with Progressive Approach for Mining Knowledge
frequent itemsets may become invalid. Thus, the maintenance
of large itemsets for dynamic databases is very costly if re-run
of previous mining algorithm on updated database is applied
because it repeats much of work done in previous
computations. Furthermore, there is not enough space to store
all the data for its processing. So instead of finding large
itemsets again some heuristics are used for mining dynamic
databases [10].
This paper is organized as follows. . In Section 2, related
work to the new algorithm is discussed In Section 3, Static
Data Mining algorithms are discussed. In Section 4, Dynamic
Data Mining algorithms are discussed. In Section 5,
progressive approach for mining is discussed. In Section 6,
results related to current work are discussed. In Section 6, the
paper is concluded.
Related Work
Static data mining algorithms like Apriori, Fp-Growth, Fast
Algorithm, Partition Based Algorithms apply only on original
database. If there is a need to modify or delete some or all the
existing set of data during the process of data mining then
repetition of whole procedure is required, which is timeconsuming in addition to its lack of efficiency. So incremental
update methods like Fast Update, Probability based &
Promising based algorithms are used to extract interesting
information from dynamic databases. On the basis of this, new
approach (PAPRIORI) can be used that takes original database
progressively i.e. read a particular set of transactions at a time
while we know the size of original database. PAPRIORI is
static data mining algorithm that uses dynamic approach.
Since execution time to generate frequent itemsets remains a
great challenge, so the goal is to calculate the execution time
of proposed approach at varying value of number of
transactions read at a time (K).
Static Data Mining
Data Mining that uses static database for mining is known as
static data mining. There are different static data mining
algorithms like Apriori, Fp-Tree, Fast algorithm, Partition
based algorithm etc.
Apriori Algorithm
Apriori is the most widely accepted static data mining
algorithm [7][9]. This is described as a “fast algorithm for
mining association rules”. Apriori algorithm is driven by
market-basket data. It efficiently generates large itemsets
along with generation of candidate itemsets by repeatedly
scanning the database. Apriori algorithm is based upon
candidate set generation and test method. The problem that
always appears during mining frequent relations is multiple
scans of original database, huge number of candidate
generation and tedious workload of support counting for
candidates. So there is need to reduce passes of transaction
database scans, to shrink number of candidates and to
facilitate support counting of candidates.
FP-Growth Algorithm
FP-Tree is an order of magnitude faster than the Apriori
algorithm. This is used for mining static databases. In this, the
frequent patterns generation process includes two sub
processes: constructing the Fp-Tree, and generating frequent
patterns from the FP tree. This uses divide-and-conquer
method and takes 2 scans of database [11]. Candidate itemsets
generation does not occur in this.
Fast Algorithm
Most time consuming operation in the discovery of association
rules from the database is the computation of the frequency of
the occurrences of interesting subset of items called
candidates. So there is need to develop a method that avoids or
reduces candidate generation and test and utilizes some novel
data structures to reduce the cost in frequent pattern mining.
Fast algorithm uses TreeMap which is a structure in java that
store key / value pair[12]. Moreover Arraylist technique that
greatly reduces the need to traverse the database is also used.
This reduces usage of memory.
Partition Based Algorithm
Partition based algorithm divides the database into partitions
that reduces the number of database scans to two. This
algorithm reduces both CPU and I/O overheads [13]. This
algorithm is especially suitable for very large size databases.
During first scan, divide database into partitions and generate
frequent itemsets in different partitions separately by scanning
the database once in each partition. During second scan,
counters for each of these itemsets are set up and their actual
support is measured to determine if they are large across entire
database. If the items are uniformly distributed across
partitions then a large fraction of itemsets will be large.
Dynamic Data Mining
Data Mining that uses dynamic databases that take into
considerations all updates (insert, update, and delete problems)
into account is known as dynamic data mining. There are
different dynamic data mining algorithms like Fast Update
(FUp), incremental method like promising based algorithm
and probability based algorithm.
Fast Update Algorithm
An incremental updating technique FUp (Fast Update)
algorithm is used for efficient maintenance of discovered
association rules when new transactional data are added to a
transaction database [14]. In this, we seperate winners (those
that remain large in updated database) from losers (that are not
large in updated database) among large items in original
database and find new winners that are large in original
database (DB) and incremental database (db) i.e. (DB U db).
This algorithm is 2 to 16 times faster than Apriori.
Promising Based Incremental Approach
Promising frequent itemset algorithm, an incremental method,
[Page No. 465]
5th IEEE International Conference on Advanced Computing & Communication Technologies [ICACCT-2011] ISBN 81-87885-03-3
Probability Based Incremental Approach
Probability-based incremental association rule discovery
algorithm is used to extract interesting information from
dynamic databases [16]. This uses principle of Bernoulli trial
to find expected frequent itemsets that reduces number of
scans to original database. This proposes a new updating and
pruning algorithm that guarantee to find all frequent itemsets
of an updated database efficiently. The results show that this
algorithm has better performance than that of FUp (Fast
Update).
New Static Data Mining Algorithm(PAPRIORI)
PAPRIORI
algorithm
generates
frequent
itemsets
progressively in static database by means of reading K
transactions at a time. It is based upon basic data mining
algorithm(Apriori). For first K transactions m large itemsets
will be generated then for next K transactions m, m+1 large
itemsets will be generated progressively and so on. This is
based on the following considerations.
• The itemsets that are counted initially or does not
satisfy minimum support are Estimated Infrequent
(EI) itemsets.
• The itemsets that satisfy minimum support threshold
are Estimated Frequent (EF) itemsets.
• CF (Confirmed Frequent) itemsets are those that have
been counted throughout whole database once and
satisfy minimum support.
• CI (Confirmed Infrequent) itemsets are those that
have been counted throughout whole database once
and do not satisfy minimum support.
• Following are the algorithmic steps:
CI.
This is repeated until Estimated Frequent (EF) and
Estimated Infrequent (EI) itemsets are present.
Experimental Setup
To evaluate the performance of PAPRIORI algorithm, the
algorithm is implemented and tested on a workstation with
Pentium(R) Dual-Core CPU, 2.19 GHz and 2.93GB main
memory. The experiments are conducted on a Synthetic
dataset and Zoo dataset. The Synthetic dataset comprises
1,000 transactions over 10 items. The Zoo dataset comprises
101 transactions over 15 items. Proposed algorithm is used to
find frequent itemsets from static database consisting of
transactions. Set fixed value of support for both datasets and
vary number of transactions read at a time (K) to calculate
execution time.
Results for Synthetic Dataset
On the basis of K and execution time the following graphs
with fixed value of support (50%, 45%) can be drawn for
analysing the results.
Execution Time of PAPRIORI at
different values of K on Support = 50 %
Execution Time
is proposed for dynamic data mining [15]. This algorithm uses
maximum support count of 1-itemsets obtained from previous
mining to estimate infrequent itemsets, called promising
itemsets, of an original database. These itemsets are capable of
being frequent itemsets when new transactions are inserted
into the original database. Thus, the algorithm reduces a
number of times to scan the original database. As a result, the
algorithm has execution time faster than that of previous
methods like FUP (Fast Update).
[Page No. 466]
PAPRIORI
Value of K Figure 1 Execution Time with Support = 50%
Execution Time of PAPRIORI at different
values of K on Support = 45 %
Execution Time
Step 1: Set all 1-itemsets as Estimated Infrequent (EI)
itemsets.
Step2: Read database with K transactions at a time (until
transactions read is less than total number of transactions in
database).
• For each transaction, increase counter for the itemset.
• For each itemset that belongs to EI if value of counter
satisfies minimum support then set itemset as EF.
• If itemsets belong to EF or CF then their immediate
superset is set as EI.
• For each itemsets that belongs to EF if it is read
throughout the whole database once move that into
CF.
• On the other hand if itemsets belongs to EI, if it is
read throughout the whole database once move it into
50
45
40
35
30
25
20
15
10
5
0
70
60
50
40
30
20
10
0
PAPRIORI
Value of K Figure 2 Execution Time with Support = 45%
Static Data Mining Algorithm with Progressive Approach for Mining Knowledge
Results for Zoo dataset
On the basis of K and execution time the following graphs
with fixed value of support (50%, 55%) can be drawn for
analysing the results.
Execution Time
Execution Time of PAPRIORI at
different values of K on Support = 50 %
50
40
30
PAPRIORI
20
10
0
20 30 40 50 60 70 80 90
Value of K Figure 3 Execution Time with Support = 50%
Execution Time
Execution Time of PAPRIORI at different
values of K on Support = 55 %
35
30
25
20
15
10
5
0
PAPRIORI
20 30 40 50 60 70 80 90
Value of K Figure 4 Execution Time with Support = 55%
It is obtained from the Figure 1, Figure 2, Figure 3 and
Figure 4 that at intermediate value of K, execution time of
PAPRIORI algorithm is less. So selection of right value of K
is required. If value of K is very less, no frequent itemsets can
be obtained easily and execution time will increase. On the
other hand, if value of K is very large then again execution
time increases and it behaves like Apriori Algorithm.
Conclusion
Mining knowledge from database is both practical and
desirable. We have proposed static data mining algorithm that
generates itemsets progressively with less execution time at
intermediate number of transactions read. In the future,
further researches and experiments on the proposed algorithm
will be presented.
References
[1] M. Dunham. “Data Mining – Introductory and
Advanced Topics”. Pg 185-186. Section 6.7.2.
Pearson Education. 2003.
[2] B.N. Lakshmi , G.H. Raghunandhan,” A Conceptual
Overview of Data Mining”, Proceedings of the
National Conference on Innovations in Emerging
Technology, pp.27-32, February 2011.
[3] Qi Luo, “Knowledge Discovery and Data Mining,” in
Proc. Workshop on Knowledge Discovery and Data
Mining, Adelaide, SA , 2008, pp 3-5,IEEE.
[4] Usama Fayyad, Gregory Piatetsky-Shapiro, and
Padhraic Smyth, “From Data Mining to Knowledge
Discovery in Databases”, American Association for
Artificial Intelligence Magazine, pp. 37-54, 1996.
[5] V.Umarani, Dr.M.Punithavalli, “A Study on Effective
Mining of Association Rules From Huge Databases”,
IJCSR International Journal of Computer Science and
Research, Vol. 1 Issue 1, 2010, pp 30-34.
[6] Jiawei Han and Micheline Kamber, “Data Mining:
Concept and Techniques,” N. Harcourt India Private
Limited ISBN: 81-7867-023-2,2nd Edition, 2001.
[7] R. Agrawal, T. Imielinski, and A. Swami, “Mining
association rules between sets of items in large
databases”. In Proceedings of the 1993 ACM
SIGMOD International Conference on Management of
Data, pages 207-216, Washington, DC, May 2628,1993.
[8] Tian Lan, Runtong Zhang and Hong Dai, “A New
Frame of Knowledge Discovery,” in Proc. 1st
International Workshop on Knowledge Discovery and
Data Mining, WKDD 2008, Jan. 2008, pp 607 – 611.
[9] Rakesh Agrawal & Ramakrishan Srikant,” Fast
algorithm for mining Association rules”, IBM
Almaden Research Center, 650 Harry road, San Jose,
CA 95120: In proceedings of the 20th VLDB
conference Santiago, Chile, pp 487-499,1994.
[10] Hebah H. O. Nasereddin, “Stream Data Mining”,
International Journal of Web Applications, Volume 1,
No. 4, December 2009, pp183-190.
[11] J. Han, J. Pei, and Y. Yin.” Mining frequent patterns
without candidate generation”, in W.Chen, J.
Naughton, and P. A.Bernstein, editors, 2000 ACM
SIGMOD Intl. Conference on Management of Data,
Vol. 29, No.2 pp 1-12.
[12] M.H.Margahny and A.A.Mitwaly,” Fast Algorithm for
Mining Association Rules”, AIML 05 Conference, pp
19-21, December 2005, CICC, Cairo, Egypt.
[13] Ashok Savasere, Edward Omiecinski, Shamkant
Navathe,” An Efficient Algorithm for Mining
Association Rules in Large Databases”, in proceedings
of 21st VLDB Conference , Zurich , Switzerland,
pp432-444, 1995.
[14] David W. Cheung, Jiawei Han, Vincent T. Ngt C.Y.
Wongj,” Maintenance of Discovered Association
Rules in Large Databases: An Incremental Updating
Technique”, in proceedings of the 12th ICDE, New
Orleans, Louisiania (IEEE) , pp 106-114,February
1996.
[Page No. 467]
5th IEEE International Conference on Advanced Computing & Communication Technologies [ICACCT-2011] ISBN 81-87885-03-3
[15] Ratchadaporn Amornchewin, Worapoj Kreesuradej,”
Incremental Association Rule Mining Using Promising
Frequent Itemset Algorithm”, 6th International
Conference on
Information, Communications &
Signal Processing ( ICICS ), 2007, IEEE, pp1-5.
[16] Ratchadaporn Amornchewin, Worapoj Kreesuradej,”
Mining Dynamic Databases using Probability-Based
Incremental Association Rule Discovery Algorithm,
Journal of Universal Computer Science, pp 24092428,Vol. 15, No.12, 28 June 2009.
[Page No. 468]