I43055257

Tania Banerjee et al Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 3( Version 5), March 2014, pp.52-57
www.ijera.com 52 | P a g e
Modified Approach for Hiding Sensitive Association Rules for
Preserving Privacy in Database
Tania Banerjee, Esha Panse, Vinay Singh, Prasanna Kharche
Department of Computer Engineering Dr. D.Y Patil College of Engineering, Ambi, Pune, India.
Abstract
Data mining is the process of analyzing large database to find useful patterns. The term pattern refers to the
items which are frequently occurring in set of transaction. The frequent patterns are used to find association
between sets of item. The efficiency of mining association rules and confidentiality of association rule is
becoming one of important area of knowledge discovery in database. This paper is organized into two sections.
In the system Apriori algorithm is being presented that efficiently generates association rules. These reduces
unnecessary database scan at time of forming frequent large item sets .We have tried to give contribution to
improved Apriori algorithm by hiding sensitive association rules which are generated by applying improved
Apriori algorithm on supermarket database. In this paper we have used novel approach that strategically
modifies few transactions in transaction database to decrease support and confidence of sensitive rule without
producing any side effects. Thus in the paper we have efficiently generated frequent item set sets by applying
Improved Apriori algorithm and generated association rules by applying minimum support and minimum
confidence and then we went one step further to identify sensitive rules and tried to hide them without any side
effects to maintain integrity of data without generating spurious rules.
KEYWORDS: Association rule, confidence, Data mining methods and Algorithm, Minimum Support
Threshold (MST), Minimum Confidence Threshold, (MCT), Rule hiding, Sensitive pattern, Sensitivity.
I. INTRODUCTION
Data mining is the process of analyzing
large database to find useful patterns. The term
pattern refers to the items which are frequently
occurring in set of transaction. The frequent patterns
are used to find association between sets of item.
Association rule mining technique is widely used in
data mining to find relationship between item sets.
The efficiency of mining association rules and
confidentiality of association rule is becoming one of
important area of knowledge discovery in data base.
However ,it breaks out many privacy issues. From a
general point of view, we may classify privacy issues
into two broad categories. The first is related to the
data perse and is known as data hiding, while the
second concerns the information ,or else the
knowledge, that a data mining method may discover
after having analyzed the data , and is known as
knowledge .Data hiding tries to remove confidential
or private information from the data before its
disclosure . It is an important aspect in improving
mining algorithm that deals with how to decrease
item sets candidate in order to generate frequent item
sets efficiently.
A. PROBLEM STATEMENT
The problem definition of our paper is
1. To hide sensitive rules without generating false
rule and display only non-sensitive rules
2. To maintain privacy in database
3. To use ISR and DSL together so as to reduce the
damage in database due to repeated sanitization.
Our paper also focuses towards making the
code more optimized and for mining association rule
we would use improved Apriori Algorithm given in
[2].
The structure of this paper is as follows:
Section II describes the literature survey and
theoretical data Section III includes MDSRRC
algorithm with some more Efficient modification in
detail with some Examples in detail Section IV
includes the comparison DSRRC over MDSRRC
Section V is conclusion
II. LITERATURE SURVEY AND
THEORETICAL DATA
Abbreviation
D Original database
D‟ Sanitized database
R Association rules generated from original
database
SR Sensitive association rules SR U R
MST Minimum support threshold
RESEARCH ARTICLE OPEN ACCESS

MCT Minimum confidence threshold
L.H.S Antecedent of an association rule
In [1], author proposed heuristic approach to
prevent disclosure of sensitive patterns. They have
mainly used two techniques : data distortion and data
blocking.[1] Data distortion is the process in
which it changes the value of an item by a new value
i.e decrease the confidence by altering „0‟ to „1‟ or
„1‟ to „0‟. [1]DATA Blocking instead of inserting or
deleting item from database it replaces „1‟ or „0‟ with
„?‟ in transaction.
Hiding sensitive rule is a subfield of privacy
preserving data mining. It is divided into two
categories- One is the preserving of data privacy, its
goal is to blur the sensitive data without changing
summary information .The other is the preserving of
information privacy, its goal is to hide the sensitive
information. A sampling approach makes a sample
database such that the sensitive rules cannot be
uncovered, while a modifying approach alters a few
parts of the database to decrease the supports or
confidences of the sensitive rules. Here, we adopt the
modifying approach since it can retain the original
data as much as possible. In the following, we focus
the discussions on hiding sensitive rule. Researcher
Atallah et al. refer to sensitive rule hiding as data
sanitization, which aims at hiding a set of sensitive
item sets, and prove that optimal sanitization is NP-
Hard. Moreover, a heuristic approach is proposed
using an item set graph to hide sensitive item sets in a
one-by-one fashion. Author Oliveira and Zaı¨ne
further address the efficiency and effectiveness
issues. They plug a transaction retrieval engine to the
hiding process and achieve a linear scalability in
terms of database size. On effectiveness, they
introduce three measures for hiding failures (sensitive
patterns that are not hidden), missing costs (non-
sensitive patterns falsely hidden), and artifactual
patterns (spurious patterns falsely generated). Since
these approaches only consider the decrease of
supports, they may fail to hide a rule if the rule can
be hidden only by decreasing the confidence.
Researcher Saygin et al. argue that both the insertion
and deletion of items will introduce false information
and make it hard to determine whether the rules
derived from the modified database can be trusted.
Therefore, they propose a modification scheme to
replace entries with unknowns. With this scheme, the
item set support is represented as a range from the
minimum support to the maximum support.
Similarly, the confidence of a rule, say X -> Y , is
also represented as a range from the minimum
confidence (the minimum support of X -> Y divided
by the maximum support of X) to the maximum
confidence (the maximum support of X -> Y divided
by the minimum support of X). By definition, a rule
X ->Y is hidden if its minimum confidence is below
MCT or the minimum support of X -> Y is below
MST. From our view, the uncertainty due to the
ranges on supports and confidences makes it hard to
determine whether the derived rules can be trusted.
Moreover, the side effects will be out of control since
they do not consider the correlation among rules in
their modification scheme. The work proposed by
author Verykios et al. in should be the most relevant
to our work. In that work, the authors propose five
algorithms for rule hiding, which are also based on
the decrease of supports and confidences. With their
assumption, their algorithms can hide one rule at a
time and decrease supports or confidences one unit at
a time. Moreover, since the authors aim at hiding all
sensitive rules instead of avoiding side effects, they
do not consider the correlation among rules in their
algorithms. Finally, to hide a rule, there can be a
large number of “candidate” entries to modify. The
“minimum impact” criterion used in the five
algorithms only focuses on minimizing the number of
modified entries. By contrast, in this system, we drop
the assumption and decide the modification schemes
and the entries to modify based on the correlation
among three kinds of rules. Thus we are using
APRIORI, MDSRRC, and ISR ,DSL n both by using
both the algorithm of isl and dsr we can save the data
base from get damage by constant sanitization.
Mining an association rule with support and
confidence is defined as follow: let I= {i1....iN } be
distinct literals called items. Given a database D=
{T1,...Tm} is a set of transaction where each
transaction T is a set of items as Ti C I (1≤i≤m). The
association rule is define as X→Y, where YCI, XCI
and Y∩X=Ø. X is called rule‟s antecedent (L.H.S)
and Y is called rule‟s consequent (R.H.S). The
support of rule X→Y is calculated using the
following formula: Support(X→Y) =|XUY|/|D|,
where |D| define the total number of the transactions
in the database D and |XUY| is the number of
transactions which support item set XY. The
confidence of rule is calculated using following
formula: Confidence (X→Y) = |XUY| / |X|, where |X|
is number of transactions which support item set X.
A rule X→Y is mined from database if support
(X→Y) ≥ MST (minimum support threshold) and
confidence (X→Y) ≥ MCT (minimum confidence
threshold).

Figure 1: flow diagram.
In the ISL, the short item set is chosen as
sacrifice item, because it has least item, negative
impact cause by modify it may be the smallest. Our
method first step is the same with ISL or DSR, then
check the mining result, in this step, using privacy
quantify to measure the result, if it's dissatisfied,
return to change the choose policy. check the whole
database, classify the items with their support, the
items with support more close to Min_sup as
"unsettled item sets" the other is "settled item sets",
when choose the sacrifice item, the item in "settled
item sets" will be chosen first.
III. THE MODIFIED ALGORITHM
The modified algorithm starts with mining
the association rule from the original database D
using association rule mining algorithm e.g. Apriori
algorithm [5]. Then user specifies some rules as
sensitive rules (SR) from the rules generated by the
association rule mining algorithm. Then algorithm
counts occurrences of each item in R.H.S of sensitive
rules. Now algorithm finds IS= {is0, is1… isk } k≤n,
by arranging those items in decreasing order of their
counts. After that sensitivity of each item is
calculated then sensitivity of each transaction is
calculated. Then transactions which support is0 are
sorted in descending order of their sensitivities. Now
rule hiding process starts by selecting first transaction
from the sorted transactions with higher sensitivity,
delete item is0 from that transaction. Then update
support and confidence of all sensitive rules and if
any rules have support and confidence below MST
and MCT respectively then delete it from SR. Then
update sensitivity of each item, transaction and IS.
Again select transaction with higher sensitivity and
delete is0 from it. This process continues until all
sensitive rules are hidden. As a result, modified
transactions are updated in the original database and
new database is generated which is called sanitized
database D‟, which preserves the privacy of sensitive
information and maintains database quality.
Proposed algorithm MDSRRC is shown
below, which is used to hide the sensitive rules from
database. Given a database D, MCT (minimum
confidence threshold) and MST (minimum support
threshold) algorithm generates sanitized database D‟.
Sanitized database hides all sensitive rules and
maintains data quality.
MDSRRC Algorithm
INPUT:
MCT (Minimum Confidence Threshold), Original
database D, MST (Minimum support threshold).
OUTPUT:
Database D‟ with all sensitive rules is
hidden.
1. Apply Apriori algorithm [3]
on given database D.
2. Generate all possible
association rules R
3. Select set of rules SR‫ؿ‬ R as
sensitive rules.
4. Calculate sensitivity of each
item j ‫א‬ D.
Calculate sensitivity of each
Transaction.
5. Count occurrences of each
item in R.H.S of sensitive
rules, find IS={is0, is1...isk}
k≤n, by arranging those items
in descending order of their
count. If two items have same
count then sort those in
descending order of their
actual support count
6. Select the transactions which
supports is0, then sort them in
descending order of their
sensitivity. If two transactions
have same sensitivity then sort
those in increasing order of
their length.
7. While(SR is not empty)
8. {
9. Start with first transaction
from sorted transactions,
10. Delete item is0 from that
transaction. 11. For each rule r
‫א‬ SR
11. {
13. Update support and
confidence of the rule r.
End

14. If(support of r < MST or
confidence of r < MCT)
15. {
16. Delete Rule r from SR.
17. Update sensitivity of
each item.
18. Update IS (This may
change is0).
19. Update the sensitivity of
each transaction
20. Select the transactions
which are supports is0,
21. Sort those in descending
order of their sensitivity.
22. }
23. Else
24. {
25. Take next transaction
from sorted transactions, go to step
10.
26. }
27. }
28. }
29. End
MDSRRC select best items so that
deleting those items hide maximum
rules from database to maintain
data quality.
Figure 2: MDSRRC algorithm in flow diagram.
Algorithm
ISL Input:
(1) a source database D,
(2) a min_support,
(3) a min_confidence,
(4) a set of predicting items X Output: a transformed
database D', where rules containing X on LHS will be
hidden
1. Find large I-item sets from D;
2. For each predicting item x Є X
3. If x is not a large I-item set, then X = X-{x};
4. If X is empty then EXIT; II no rule contains X in
LHS
5. Find large 2-item sets from D;
6. For each x Є X {
7. For each large 2-itemset containing x {
8. Compute confidence of rule U, where U is a rule
like x-y;
9. If conf{U} < min_con , then
10. Go to next large 2-itemset;
11. Else {Increase Support of LHS
12. Find T L = {t in Dlt does not support U};
13. Sort TL in ascending order by the number of
items;
14. While {conf(U ) '?:. min _ conf and TL is not
empty }{
15. Choose the first transaction t from TL;
16. Modify t to support x, the LHS(U );
17. Compute support and confidence of U;
18. Remove and save the first transaction t from TL;
19. }; II end While
20. }; II end if conf(U) < min_cmif
21. If TL is empty, then {
22. Cannot hide x-y;
23. Restore D;
24. Go to next large-2 itemset;
25. } II end if TL is empty
26. } II end of for each large 2-itemset
27. Remove x from X;
28. } II end of for each xE X
29. Output updated D, as the transformed D'
A. EXAMPLES
Example 1: Let a bag store that purchase luggage
bag from two companies, XYZ and PQR, and both
can access customers‟ database of the store. Now
XYZ applies data mining techniques and mines
association rules related to PQR‟s products. XYZ had
found that most of the customer who buy luggage bag
of the PQR also buy college bag. Now XYZ offers
some discount on college bag if customer purchases
XYZ‟s luggage bag. As result the business of PQR
goes down. So releasing the database with sensitive
information can cause the problem. This scenario
gives the direction to research on sensitive rules (or
knowledge) hiding in database.
Example 2:

The support of an association pattern is the
percentage of task-relevant data transactions for
which the pattern
is true.
Support(A => B) = # tuples containing both A and B
total # of tuples
Confidence is defined as the measure of
certainty or trustworthiness associated with each
discovered
pattern.
confidence (A=> B)= # tuples containing both A and B
# tuples containing A
Item
ID
Item Suport
I1 Tea 4/8
I2 Sugar 3/8
I3 Toothbrush 2/8
I4 Toothpaste 2/8
I5 Pen 1/8
I6 Refill 1/8
I7 Maggie 1/8
I8 Coffee 2/8
Id Item Support
I1 Tea 4/8
I2 Sugar 3/8
Min_sup = 37.5%(3/8)
IV. COMPARISON OF MDSRRC AND
DSRRC
Modified Decrease Support of R.H.S Item of
Rule Clusters (MDSRRC) is an improved version of
Decrease Support of R.H.S Item of Rule Clusters.
This algorithm selects the items and transactions
based on certain criteria which modify transactions to
hide the sensitive information. DSRRC could not
hide association rules with multiple items in
antecedent (L.H.S) and consequent (R.H.S). To
overcome this limitation, we have proposed an
algorithm MSDRRC which uses count of items in
consequent of the sensitive rules. MDSRRC modifies
the minimum number of transaction to hide
maximum sensitive rules and maintain data quality.
In the modified algorithm, we used heuristic
approaches for hiding the sensitive rules these
approaches are data distortion and data blocking. In
data distortion we changed the item value with a new
value, means it alter '0' to '1' or '1' to '0' for selected
items to decrease the confidence. In data blocking
instead of inserting or deleting items from the data
base it simply replaces '1' and '0' with '?' in selected
transactions.
After applying both Algorithms on sample
database we have done evaluation by considering the
performance parameters which are given in [13] viz.
(a) HF (hiding failure): It is the percentage of the
sensitive data that remain exposed in the sanitized
dataset. (b) MC (misses cost): It is the percentage of
the non-sensitive data that are hidden as a side- effect
of the sanitization process. (c) AP (artificial patterns):
It is the percentage of the discovered patterns that are
artifacts. (d) DISS (dissimilarity): It is the difference
between the original and the sanitized datasets. (e)
SEF (side effect factor): It is the amount of non-
sensitive association rules that are removed as an
effect of the sanitization process. Experimental
results show that MDSRRC increase efficiency and
reduce modification of transactions in database.
Performance comparison of MDSRRC with
algorithm DSRRC is given in Table VI.
Table VI. Performance result
Fig. 2. Performance Comparison between DSRRC,
MDSRRC
TID List of Items
1 Tea(I1),Sugar(I2),Toothbrush(I3),Toothpaste(I
4),Pen(I5),Refill(I6)
2 Toothbrush(I3),Toothpaste(I4)
3 Tea(I1),Sugar(I2),Maggie(I7)
4 Sugar(I2), Tea(I1),Coffee(I8)
5 Tea(I1), Coffee(I8)

V. CONCLUSION
We proposed an algorithm named MDSRRC
which hides sensitive association rules on data base
to maintain data quality. The algorithm MDSRRC is
an improved version of DSRRC .MDSRRC modifies
minimum no transaction to hide maximum sensitive
rules and improved data quality. We have used an
improved Apriori algorithm which is used in data
mining to extract data. The improved Apriori takes
less time for generating frequent item sets. In future
MDSRRC can be used further to increase the
efficiency and reduce the side effects by minimizing
modification on data base. In future the improved
Apriori algorithm will be more efficient to generate
strong rules while hiding the sensitive rules.
REFERENCES
[1] Nikunj H. Domadiya and Udai Pratap Rao,
“Hiding Sensitive Association Rules to
Maintain Privacy and Data Quality in
Database”3rd IEEE International Advance
Computing Conference (IACC) 2013.
[2] V. S. Verykios, A. K. Elmagarmid, E.
Bertino, Y. Saygin, and E. Dasseni,
“Association rule hiding,” IEEE
Transactions on Knowledge and Data
Engineering, vol. 16, pp. 434–447, 2004.
[3] C. N. Modi, U. P. Rao, and D. R. Patel,
“Maintaining privacy and data quality in
privacy preserving association rule mining,”
2010 Second International conference on
Computing, Communication and
Networking Technologies, pp. 1–6, Jul.
2010.
[4] J. Han, Data Mining: Concepts and
Techniques. San Francisco, CA, USA:
Morgan Kaufmann Publishers Inc., 2005
[5] Y.-H. Wu, C.-M. Chiang, and A. L. Chen,
“Hiding sensitive association rules with
limited side effects,” IEEE Transactions on
Knowledge and Data Engineering, vol. 19,
pp. 29–42, 2007.
[6] S.-L. Wang, B. Parikh, and A. Jafari,
“Hiding informative association rule sets,”
Expert Systems with Applications, vol. 33,
no. 2, pp. 316 – 323, 2007.
[7] S.-L.Wang, D. Patel, A. Jafari, and T.-P.
Hong, “Hiding collaborative
recommendation association rules,” Applied
Intelligence, vol. 27, pp. 67–77, 2007.
[8] D.F. Gleich and L.-H. Lim. Rank
aggregation via nuclear norm minimization.
In KDD, 2011.
[9] Y. Saygin, V. S. Verykios, and A. K.
Elmagarmid, “Privacy preserving
association rule mining.” in RIDE. IEEE
Computer Society, 2002, pp 151–158.
[10] C. N. Modi, U. P. Rao, and D. R. Patel, “An
Efficient Solution for Privacy Preserving
Association Rule Mining,” (IJCNS)
International Journal of Computer and
Network Security, vol. 2, no. 5, pp. 79–85,
2010.
[11] Wu and H. Wang, “Research on the privacy
preserving algorithm of association rule
mining in centralized database,” in
Proceedings of the 2008 International
Symposiums on Information Processing, ser.
ISIP‟08. Washington, DC, USA: IEEE
Computer Society, 2008, pp. 131–134
[12] V. Verykios and A. Gkoulalas-Divanis, A
Survey of Association Rule Hiding Methods
for Privacy, ser. Advances in Database
Systems, C. [13] M. Atallah, A.
Elmagarmid, M. Ibrahim, E. Bertino, and V.
Verykios, “Disclosure limitation of sensitive
rules,” in Proceedings of the 1999
Workshop on Knowledge and Data
Engineering Exchange, ser. KDEX ‟99.
Washington, DC, USA: IEEE Computer
Society, 1999, pp.

I43055257

More Related Content

What's hot (17)

Viewers also liked (20)

Similar to I43055257 (20)

Recently uploaded (20)

I43055257