![International Journal on Cybernetics & Informatics ( IJCI) Vol. 3, No. 1, February 2014
2
preserving approaches for microdata publishing. Section 6, deals with examples of several
applications that ensure privacy. Section 7, discuss the possible privacy threats on microdata
publishing. Section 8, present the study on various privacy preserving techniques. Section 9,
introduce the comparative study on privacy preserving techniques.
2. NEED FOR PRIVACY PRESERVATION
Privacy preservation is considered as an important factor for effective utilization of the massive
volume of data. This data have been stored in electronic format, without disturbing an individual.
It preserves privacy during data collection and mining.
3. DATA COLLECTION AND MINING: Why and What:
In this information era, collected data and its transactions are recorded somewhere. Many data
security-enhanced techniques have been developed to secure the privacy of the individual while
collecting and mining the data. There is a demand to exchange and publish data
among various parties after collecting it. The data would be collected from different sources and
they cannot be shared directly. The owner might collect data and process it through various
anonymization techniques in order to mine the data and release it to the recipient. The collected
data would be prepared well before publishing in order to ensure confidentiality and privacy.
Fig 1: A Simple privacy model
4. MICRODATA PUBLISHING
Privacy preserving data publishing contains microdata. It provides information about an
individual. The record holds information in the form of:
Identifier (I), Quasi Identifier (QI), Sensitive Attribute (SA). The Identifiers are explicitly known
only to the data holder, Quasi Identifiers which are non sensitive attributes and sensitive Attribute
contains individual’s specific information that was unknown to an adversary.
5. PRIVACY PRESERVING APPROACHES
According to Sweeny [9], the specific information about a person are stored in the table (or a
relation) of columns (or attributes) and rows (or records). Several privacy preserving approaches
in data publishing such as randomization, sampling, cell suppression, data swapping and
perturbation have been designed for microdata publishing [10]. Privacy preserving crossed over
DATA PUBLISHER DATA RECIPIENT
ANONYMIZATIONORIGINAL
DATASET
RELEASED
DATASET](https://image.slidesharecdn.com/asurveyonprivacypreservingdatapublishing-140312014553-phpapp01/85/A-survey-on-privacy-preserving-data-publishing-2-320.jpg)
![International Journal on Cybernetics & Informatics ( IJCI) Vol. 3, No. 1, February 2014
3
various stages in its development. Due to the level of complexity in the existing technique, a
thought to preserve privacy considered as a new area of research.
5.1 Randomization
Randomization was an ability to anonymize the entire dataset, to preserve certain semantics. In
the existing privacy preserving data mining techniques, the randomization is considered as one of
the most important technique. This provides knowledge discovery and a balance between privacy
and utility [11]. The balance between privacy and utility was achieved by adding noise to the
data. The randomized data after balanced would transmit to the recipient. The recipient would
receive the data using distribution reconstruction algorithm.
5.2 Suppression
Suppression have not involved in releasing the actual value of the data. Suppression
would replace the value of specific attribute description, typically the Quasi Identifiers as the attr
ibutes, with less specific description. Like generalization, Suppression would hide some details of
Quasi Identifiers [13].A specific value could be replaced by a generic value at the time of
suppression. Suppression indicates that replaced value was not disclosed [13].
6. PRIVACY PRESERVING APPLICATIONS
6.1 Medical Database: Scrub System
As per the author perceptions [16], clinical information would be in the form of text, which
contains information of patients like his family members, address, blood group and phone
number. Traditional technique have been used only for global search and replace procedure in
order to maintain privacy [16].As per author Sweeny L findings, Scrub system used numerous
detection algorithms in order to support the privacy.
6.2 Bioterrorism Application
It is essential to analyse the medical data for privacy preservation in the bioterrorism
application. For example, Biological agents are widely found in the natural environment
such as anthrax. It is important to find the anthrax attack from the normal attack [17].It is
necessary to track incidences of the common diseases. The corresponding data would be reported
to the public health agencies. The respiratory diseases were not reportable-diseases. This provides
a solution for more identifiable information in accordance with public health law [17].
7. PRIVACY THREATS
Releasing the result of data mining could cause privacy threats. Several privacy
disclosure threats were possible in microdata publishing like identity disclosure, membership
disclosure and an attribute disclosure. Privacy threats results in more disclosure risk.
Anonymizing the data and preserving the data through various disclosure protections would result
in better utility.
7.1 Identity Disclosure
Usually an individual was linked to a record in the published table. If his identity was disclosed,
then the corresponding sensitive value of an individual would be revealed](https://image.slidesharecdn.com/asurveyonprivacypreservingdatapublishing-140312014553-phpapp01/85/A-survey-on-privacy-preserving-data-publishing-3-320.jpg)
![International Journal on Cybernetics & Informatics ( IJCI) Vol. 3, No. 1, February 2014
4
7.2 Attribute Disclosure
Attribute disclosure was possible when information about individual record would be revealed.
Before releasing the data, it is the must to infer attributes of an individual with high confidence.
As per the authors view [19], matching multiple bucket was important to protect attribute
disclosure.
7.3 Membership Disclosure
Membership information in the released table would infer an identity of an individual through
various attacks. If the selection criteria were not a sensitive attribute value, then it would lead to
have a membership disclosure [18].
8. RELATED WORK
8.1 K-Anonymity
The various phenomena arise when analysing and publishing the data in high-dimensional space.
K-Anonymity was a method to support the curse [1]. Generalization on K-Anonymity was
applied to mask the exact value of an attribute [1]. The perturbation technique on K-Anonymity
was suitable for aggregate distribution of an individual than the inter-attribute relation of an
individual. 2-anonymity and Gaussian cluster methods proposed on K-Anonymity technique,
ensure privacy by evaluating probability and assigning its value to zero [1]. As per the authors
view [1], this method tried to understand the probability distribution which would have maximum
likelihood of its attributes. As per the authors [1], there would be a loss for high-dimensional data.
8.2 ℓ-Diversity
Information about an individual could not be published without revealing sensitive attribute [2].
K-Anonymization was not enough to protect the data which include homogeneity attack and
background knowledge attack [2].ℓ-Diversity technique described that sensitive attributes would
have at most same frequency. For example, with positive disclosure, if Alice wants to discover
Bob, Alice would determine Bob with high-probability distribution. The negative disclosure
would happen when an adversary could correctly eliminate some possible value of the sensitive
attributes .There could be a minimum difference between the prior belief and posterior belief [2].
8.3 t-closeness
Anil Prakash, Ravindar Mogili found that K-Anonymity and ℓ -Diversity was not used to prevent
attribute disclosure [3]. ℓ-Diversity would have well represented sensitive attribute value that
was assigned only with certain number of limitations [2]. t-closeness has been proposed to
describe the distribution of sensitive attribute with equivalence class. Earth Mover Distance was
used to measure the distance between the two probabilistic distributions [3].conjunction has been
proposed to combine machine learning and statistical analysis. Closeness among the columns was
reduced by aggregation [3].](https://image.slidesharecdn.com/asurveyonprivacypreservingdatapublishing-140312014553-phpapp01/85/A-survey-on-privacy-preserving-data-publishing-4-320.jpg)
![International Journal on Cybernetics & Informatics ( IJCI) Vol. 3, No. 1, February 2014
5
8.4 Km
Anonymity
Km
Anonymity has been proposed for an anonymize transactional database [4].
Km
Anonymity aim at protect the database against an adversary who has knowledge about almost
m items in the transaction [18]. The generalization was used to maintain the set valued data. For
any transaction on K-1 records, other identical transaction would also appear. Km
anonymity has
been introduced via top down local generalization process to record the number of transaction
records [4]. The partition based approach was used to group (partition) the similar items in a top-
down manner [4]. The km
anonymity model would help to prevent privacy breaches raised from
an adversary who would discovered m items in a transaction database.
8.5 Distributed K-Anonymity framework (DKA)
The collection of data from different sites cannot be shared directly. The key step was to
anonymize the data in order to generalise a specific value [5]. A secure 2-party framework was
designed for multiparty computation that has been used to join the dataset from various sites [5].
Distributed K-Anonymity (DKA) prevent identification of an individual by make use of global
Anonymization in the encrypted form. DKA provide a secure framework between two parties [5].
Two parties would agree on Global Anonymization algorithm that could produce local K-
Anonymous dataset. In addition, DKA provide a secure distributed protocol which would require
that two parties could mutually semi-honest [5]. Still the trade-off between utility and potential of
data was misused in DKA [5].
8.6 K-Anonymity Clustering
Among various clustering methods, hierarichal clustering was mostly used to achieve K-
Anonymity [14]. Weighted Feature C-means Clustering [WFC] used to reduce the information
distortion. WFC partition all records into equivalence class and would merge the class using class
merging mechanism [6]. The numerical values of quasi identifier were used to evaluate the
Weighted Feature C-means Clustering technique. The authors also try to provide the
dissimilarity evaluated approach which would take different types of feature values for class
merging mechanism.
8.7 R-U Confidentiality Map
Normally, Anonymization would cause loss in potential utility gain. The most important factor
that should be balanced are Risk(R) and Utility (U). The fundamental characteristic of
generalization and bucketization was compared [8]. Privacy trade-off
utility have fixed the privacy requirement with various privacy parameters and produceda
n anonymized dataset. Generalization would work on three methods such as Apriori
Anonymization (AA), the Constrained Based Anonymization Transaction (COAT) and
Privacy-Constrained cluster Based Transaction Algorithm (PCTA).These three methods were
used to maintain the association between Risk(R)-Utility (U). The trade-off was gained by
combining the above three methods as (AA & COAT) and ( COAT & PCTA) [8]. It was essential
to maintain the histogram H for the occurrence of each sensitive item in the transaction database
[8].
As per the authors [8], protection of the data requires:
(i) Security, to ensure that there was no loss of the stored information.
(ii) Confidentiality, to ensure that no one can drop data while the data were transmitted
between authorized users.](https://image.slidesharecdn.com/asurveyonprivacypreservingdatapublishing-140312014553-phpapp01/85/A-survey-on-privacy-preserving-data-publishing-5-320.jpg)
![International Journal on Cybernetics & Informatics ( IJCI) Vol. 3, No. 1, February 2014
6
(iii) Anonymity, to ensure private and sensitive information about an individual was not
disclosed when a record was released.
8.8 Slicing
K-Anonymity could not guarantee the background knowledge attack of an adversary [1]. The data
was partitioned in both horizontal and vertical direction to preserve the privacy by permuting the
sensitive attribute [8].Slicing was a popular data Anonymization technique. Slicing could be
formalized by comparing with generalization and bucketization [18]. Generalization and
Bucketization were used to replace a specific sensitive attribute value of transaction records [19].
Generalization changed the low conceptual level of data to high conceptual level. Bucketization
used l-diversity checking technique for multiple matched buckets. Mondrian algorithm [18]
support membership disclosure protection for different privacy threats. Privacy threats leads to
identity disclosure, attribute disclosure and membership disclosure. Mondrian algorithm would
support map between logical model and a physical model. Slicing support prevention on
membership disclosure but still risk and utility was not achieved. Slicing preserve correlation on
attributes. A tuple could potentially match with multiple buckets to avoid privacy threats with
slicing [18]. Slicing supported data utility than the traditional techniques.
8.9 Overlapped Slicing
The problem was broken down into sub problems, which were reused several times. Overlapped
slicing would duplicate an attribute with more than one column. Each column results in more
attribute correlations [19].Horizontal and vertical partition was done by duplicating the attributes
in more than one columns. Tuples were grouped together in horizontal partitioning. Attributes
were correlated in vertical partitioning. Sensitive attribute would be placed in each column.
Random permutation would take place on Sensitive Attribute [SA] column [19].The authors tried
to provide protection against membership disclosure by distinguishing the original tuples from the
fake tuples where the disclosure risk has been occurred. The matching stratergy in the overlapped
slicing table is low. Overlapped slicing enhanced with slicing provided the membership
disclosure protection. But, it leads to more attribute correlations and there would be a secrecy loss
of privacy in some extent.](https://image.slidesharecdn.com/asurveyonprivacypreservingdatapublishing-140312014553-phpapp01/85/A-survey-on-privacy-preserving-data-publishing-6-320.jpg)
![International Journal on Cybernetics & Informatics ( IJCI) Vol. 3, No. 1, February 2014
8
REFERENCES
[1] Charu C. Aggarwal, (2005), ‘‘On k-Anonymity and the Curse of Dimensionality”, Proceedings
of the 31st VLDB Conference, Trondheim, Norway, pp.901-909
[2] Ashwin Machanavajjhala , Daniel Kifer,Johannes Gehrke, Muthuramakrishnan Venkita
Subramanian, (2006),“ ℓ-Diversity : Privacy Beyond K-Anonymity”, Proc.International
conference on Data Engineering.(ICDE),pp.24.
[3] Anil Prakash, Ravindar Mogili ,(2012),‘‘Privacy Preservation Measure using t-closeness with
combined l-diversity and k-anonymity”, International Journal of Advanced Research in Computer
Science and Electronics Engineering (IJARC SEE)Volume 1, Issue 8,pp:28-33
[4] Yeye He, Jeffery Naughton .F, (2009), “Anonymization of Set Valued Data via Top Down
Local Generalization”, Proc. International Conference on Very Large Databases (VLDB), pp.934-
945.
[5] Wei Jiang , Chris Clifton, (2006)‘‘ A secure distributed framework for achieving k-
anonymity”, the VLDB Journal , Vol.15, No.4, pp.316-333
[6] Chuang-Cheng Chiu , ChiehYuan Tsai, (2007),“ A k Anonymity Clustering method for
Effective Data Privacy Preservation”, Springer journal on Verlag Berlin Heidelberg , pp.88-99.
[7] Grigorios Loukides, Aris Gkoulalas - Divanis, and Jianhua Shao, (2012), ‘‘Assessing Disclosure
Risk and Data Utility Trade-off in Transaction Data Anonymization”, International Journal of
Software and Informatics, Vol.6, No. 3, pp.359-361
[8] Ravindra S, Wanjari Prof .Devi,(2013), “Improving the implementation of new approach for Data
Privacy Preserving in Data Mining using slicing”. International Journal of Modern Engineering
Research (IJMER), Vol. 3, Issue. 3.
[9] L. Sweeney, (2002) “k-anonymity: a model for protecting privacy”, International Journal on
Uncertainty, Fuzziness and Knowledge based Systems, pp. 557-570.
[10] Yan Zhao, Ming Du, Jiajin Le, Yongcheng Luo,(2009), “ A Survey on Privacy Preserving
Approaches in Data Publishing” in the First International Workshop on Database Technology and
Applications
[11] Mohnish Patel, Prashant Richariya, Anurag Shrivastava, (2013),‘‘A review paper on Privacy-
Preserving Data Mining”, Review article on Scholars Journal of Engineering and Technology
(SJET) , pp.359-361
[12] Agarwa, Srikan R., (2000) ‘‘Privacy Preserving Data Mining”, In Proc. ACM SIGMO, conference
on management of data (SIGMOD’00), Dallas, TX,pp.439-450.
[13] Benjamin c.m, Fung, ke wang, rui chen, philips s.yu ,(2010),‘‘Privacy Preserving Data
Publishing:A Survey of Recent Development ”ACM Computing surveys, Vol.42, No.4,
pp.523-553
[14] Byun, J.W, Kamra, A, Bertino, Li, N, (2007), ‘‘Efficient k-Anonymization Using clustering
Techniques”. International Conference on Database Systems for Advanced Applications ,
pp.188-200.
[15] N. Li, T. Li, and S. VenkataSubramanian, (2007), “ t-closeness : Privacy beyond k-anonymity
and L -diversity,” Proc. International Conference on Data Engineering (ICDE), pp.106-115.
[16] Sweeney L, (1996), “Replacing Personally Identifiable Information in Medical Records, the
Scrub System”. Journal of the American Medical Informatics Association.
[17] Charu C.Aggarwal, “A General survey of privacy preserving Data Mining Models and Algorithms”,
IBM,T. J. Watson Research Centre
[18] Tiancheng Li , Jian Zhang , Ian Molloy ,(2012),“Slicing: A New Approach for Privacy
Preserving Data Publishing” IEEE Transaction on KDD.
[19] B.Vani, D.Jayanthi, (2013), “Efficient Approach for Privacy Preserving Microdata Publishing
Using Slicing” IJRCTT.](https://image.slidesharecdn.com/asurveyonprivacypreservingdatapublishing-140312014553-phpapp01/85/A-survey-on-privacy-preserving-data-publishing-8-320.jpg)
A survey on privacy preserving data publishing
- 1. International Journal on Cybernetics & Informatics ( IJCI) Vol. 3, No. 1, February 2014 DOI: 10.5121/ijci.2014.3101 1 A SURVEY ON PRIVACY PRESERVING DATA PUBLISHING S.Gokila1 , Dr.P.Venkateswari2 1 Computer Science and Engineering, Erode Sengunthar Engineering College, Anna University Chennai, Tamilnadu 2 Computer Science and Engineering, Erode Sengunthar Engineering College, Anna University Chennai, Tamilnadu ABSTRACT Data mining is a computational process of analysing and extracting the data from large useful datasets. In recent years, exchanging and publishing data has been common for their wealth of opportunities. Security, Privacy and data integrity are considered as challenging problems in data mining.Privacy is necessary to protect people’s interest in competitive situations. Privacy is an abilityto create and maintain different sort of social relationships with people. Privacy Preservation is one of the most important factor for an individual since he should not embarrassed by an adversary. The Privacy Preservation is an important aspect of data mining to ensure the privacy by various methods. Privacy Preservation is necessary to protect sensitive information associated with individual. This paper provides a survey of key to success and an approach where individual’s privacy would to be non-distracted. KEYWORDS Data mining, Privacy, Privacy Preserving 1. INTRODUCTION Data mining is a multidisciplinary field, would drawn from areas including database technology, machine learning, information retrieval, knowledge-based system, high performance computing and data visualization. Data mining has attracted a great deal of attention in the industry to maintain information. Due to the wide availability of data and imminent need, data are turned into useful information and knowledge. The information and knowledge gained could be used for various applications ranging from market analysis, fraud detection and customer retention. Due to the widespread of data, data mining has been viewed as a threat to privacy that are maintained by the industry. Data mining incorporate privacy as a functional component for the gained information and knowledge. Preservation of individual’s information is an essential for the data owners to ensure his privacy. Privacy plays an important role in data publishing. Data mining process allows a company to use large amount of data to develop correlations and relationships among the data to improve the business efficiency. Therefore Privacy preserving data mining has become important field of research. This paper is organised as follows. Section 2, introduce the need for privacy preservation. Section 3, summarize the importance of data collection. Section 4, discuss about various methods for partitioning of data collection for publishing microdata. Section 5, summarize various privacy
- 2. International Journal on Cybernetics & Informatics ( IJCI) Vol. 3, No. 1, February 2014 2 preserving approaches for microdata publishing. Section 6, deals with examples of several applications that ensure privacy. Section 7, discuss the possible privacy threats on microdata publishing. Section 8, present the study on various privacy preserving techniques. Section 9, introduce the comparative study on privacy preserving techniques. 2. NEED FOR PRIVACY PRESERVATION Privacy preservation is considered as an important factor for effective utilization of the massive volume of data. This data have been stored in electronic format, without disturbing an individual. It preserves privacy during data collection and mining. 3. DATA COLLECTION AND MINING: Why and What: In this information era, collected data and its transactions are recorded somewhere. Many data security-enhanced techniques have been developed to secure the privacy of the individual while collecting and mining the data. There is a demand to exchange and publish data among various parties after collecting it. The data would be collected from different sources and they cannot be shared directly. The owner might collect data and process it through various anonymization techniques in order to mine the data and release it to the recipient. The collected data would be prepared well before publishing in order to ensure confidentiality and privacy. Fig 1: A Simple privacy model 4. MICRODATA PUBLISHING Privacy preserving data publishing contains microdata. It provides information about an individual. The record holds information in the form of: Identifier (I), Quasi Identifier (QI), Sensitive Attribute (SA). The Identifiers are explicitly known only to the data holder, Quasi Identifiers which are non sensitive attributes and sensitive Attribute contains individual’s specific information that was unknown to an adversary. 5. PRIVACY PRESERVING APPROACHES According to Sweeny [9], the specific information about a person are stored in the table (or a relation) of columns (or attributes) and rows (or records). Several privacy preserving approaches in data publishing such as randomization, sampling, cell suppression, data swapping and perturbation have been designed for microdata publishing [10]. Privacy preserving crossed over DATA PUBLISHER DATA RECIPIENT ANONYMIZATIONORIGINAL DATASET RELEASED DATASET
- 3. International Journal on Cybernetics & Informatics ( IJCI) Vol. 3, No. 1, February 2014 3 various stages in its development. Due to the level of complexity in the existing technique, a thought to preserve privacy considered as a new area of research. 5.1 Randomization Randomization was an ability to anonymize the entire dataset, to preserve certain semantics. In the existing privacy preserving data mining techniques, the randomization is considered as one of the most important technique. This provides knowledge discovery and a balance between privacy and utility [11]. The balance between privacy and utility was achieved by adding noise to the data. The randomized data after balanced would transmit to the recipient. The recipient would receive the data using distribution reconstruction algorithm. 5.2 Suppression Suppression have not involved in releasing the actual value of the data. Suppression would replace the value of specific attribute description, typically the Quasi Identifiers as the attr ibutes, with less specific description. Like generalization, Suppression would hide some details of Quasi Identifiers [13].A specific value could be replaced by a generic value at the time of suppression. Suppression indicates that replaced value was not disclosed [13]. 6. PRIVACY PRESERVING APPLICATIONS 6.1 Medical Database: Scrub System As per the author perceptions [16], clinical information would be in the form of text, which contains information of patients like his family members, address, blood group and phone number. Traditional technique have been used only for global search and replace procedure in order to maintain privacy [16].As per author Sweeny L findings, Scrub system used numerous detection algorithms in order to support the privacy. 6.2 Bioterrorism Application It is essential to analyse the medical data for privacy preservation in the bioterrorism application. For example, Biological agents are widely found in the natural environment such as anthrax. It is important to find the anthrax attack from the normal attack [17].It is necessary to track incidences of the common diseases. The corresponding data would be reported to the public health agencies. The respiratory diseases were not reportable-diseases. This provides a solution for more identifiable information in accordance with public health law [17]. 7. PRIVACY THREATS Releasing the result of data mining could cause privacy threats. Several privacy disclosure threats were possible in microdata publishing like identity disclosure, membership disclosure and an attribute disclosure. Privacy threats results in more disclosure risk. Anonymizing the data and preserving the data through various disclosure protections would result in better utility. 7.1 Identity Disclosure Usually an individual was linked to a record in the published table. If his identity was disclosed, then the corresponding sensitive value of an individual would be revealed
- 4. International Journal on Cybernetics & Informatics ( IJCI) Vol. 3, No. 1, February 2014 4 7.2 Attribute Disclosure Attribute disclosure was possible when information about individual record would be revealed. Before releasing the data, it is the must to infer attributes of an individual with high confidence. As per the authors view [19], matching multiple bucket was important to protect attribute disclosure. 7.3 Membership Disclosure Membership information in the released table would infer an identity of an individual through various attacks. If the selection criteria were not a sensitive attribute value, then it would lead to have a membership disclosure [18]. 8. RELATED WORK 8.1 K-Anonymity The various phenomena arise when analysing and publishing the data in high-dimensional space. K-Anonymity was a method to support the curse [1]. Generalization on K-Anonymity was applied to mask the exact value of an attribute [1]. The perturbation technique on K-Anonymity was suitable for aggregate distribution of an individual than the inter-attribute relation of an individual. 2-anonymity and Gaussian cluster methods proposed on K-Anonymity technique, ensure privacy by evaluating probability and assigning its value to zero [1]. As per the authors view [1], this method tried to understand the probability distribution which would have maximum likelihood of its attributes. As per the authors [1], there would be a loss for high-dimensional data. 8.2 ℓ-Diversity Information about an individual could not be published without revealing sensitive attribute [2]. K-Anonymization was not enough to protect the data which include homogeneity attack and background knowledge attack [2].ℓ-Diversity technique described that sensitive attributes would have at most same frequency. For example, with positive disclosure, if Alice wants to discover Bob, Alice would determine Bob with high-probability distribution. The negative disclosure would happen when an adversary could correctly eliminate some possible value of the sensitive attributes .There could be a minimum difference between the prior belief and posterior belief [2]. 8.3 t-closeness Anil Prakash, Ravindar Mogili found that K-Anonymity and ℓ -Diversity was not used to prevent attribute disclosure [3]. ℓ-Diversity would have well represented sensitive attribute value that was assigned only with certain number of limitations [2]. t-closeness has been proposed to describe the distribution of sensitive attribute with equivalence class. Earth Mover Distance was used to measure the distance between the two probabilistic distributions [3].conjunction has been proposed to combine machine learning and statistical analysis. Closeness among the columns was reduced by aggregation [3].
- 5. International Journal on Cybernetics & Informatics ( IJCI) Vol. 3, No. 1, February 2014 5 8.4 Km Anonymity Km Anonymity has been proposed for an anonymize transactional database [4]. Km Anonymity aim at protect the database against an adversary who has knowledge about almost m items in the transaction [18]. The generalization was used to maintain the set valued data. For any transaction on K-1 records, other identical transaction would also appear. Km anonymity has been introduced via top down local generalization process to record the number of transaction records [4]. The partition based approach was used to group (partition) the similar items in a top- down manner [4]. The km anonymity model would help to prevent privacy breaches raised from an adversary who would discovered m items in a transaction database. 8.5 Distributed K-Anonymity framework (DKA) The collection of data from different sites cannot be shared directly. The key step was to anonymize the data in order to generalise a specific value [5]. A secure 2-party framework was designed for multiparty computation that has been used to join the dataset from various sites [5]. Distributed K-Anonymity (DKA) prevent identification of an individual by make use of global Anonymization in the encrypted form. DKA provide a secure framework between two parties [5]. Two parties would agree on Global Anonymization algorithm that could produce local K- Anonymous dataset. In addition, DKA provide a secure distributed protocol which would require that two parties could mutually semi-honest [5]. Still the trade-off between utility and potential of data was misused in DKA [5]. 8.6 K-Anonymity Clustering Among various clustering methods, hierarichal clustering was mostly used to achieve K- Anonymity [14]. Weighted Feature C-means Clustering [WFC] used to reduce the information distortion. WFC partition all records into equivalence class and would merge the class using class merging mechanism [6]. The numerical values of quasi identifier were used to evaluate the Weighted Feature C-means Clustering technique. The authors also try to provide the dissimilarity evaluated approach which would take different types of feature values for class merging mechanism. 8.7 R-U Confidentiality Map Normally, Anonymization would cause loss in potential utility gain. The most important factor that should be balanced are Risk(R) and Utility (U). The fundamental characteristic of generalization and bucketization was compared [8]. Privacy trade-off utility have fixed the privacy requirement with various privacy parameters and produceda n anonymized dataset. Generalization would work on three methods such as Apriori Anonymization (AA), the Constrained Based Anonymization Transaction (COAT) and Privacy-Constrained cluster Based Transaction Algorithm (PCTA).These three methods were used to maintain the association between Risk(R)-Utility (U). The trade-off was gained by combining the above three methods as (AA & COAT) and ( COAT & PCTA) [8]. It was essential to maintain the histogram H for the occurrence of each sensitive item in the transaction database [8]. As per the authors [8], protection of the data requires: (i) Security, to ensure that there was no loss of the stored information. (ii) Confidentiality, to ensure that no one can drop data while the data were transmitted between authorized users.
- 6. International Journal on Cybernetics & Informatics ( IJCI) Vol. 3, No. 1, February 2014 6 (iii) Anonymity, to ensure private and sensitive information about an individual was not disclosed when a record was released. 8.8 Slicing K-Anonymity could not guarantee the background knowledge attack of an adversary [1]. The data was partitioned in both horizontal and vertical direction to preserve the privacy by permuting the sensitive attribute [8].Slicing was a popular data Anonymization technique. Slicing could be formalized by comparing with generalization and bucketization [18]. Generalization and Bucketization were used to replace a specific sensitive attribute value of transaction records [19]. Generalization changed the low conceptual level of data to high conceptual level. Bucketization used l-diversity checking technique for multiple matched buckets. Mondrian algorithm [18] support membership disclosure protection for different privacy threats. Privacy threats leads to identity disclosure, attribute disclosure and membership disclosure. Mondrian algorithm would support map between logical model and a physical model. Slicing support prevention on membership disclosure but still risk and utility was not achieved. Slicing preserve correlation on attributes. A tuple could potentially match with multiple buckets to avoid privacy threats with slicing [18]. Slicing supported data utility than the traditional techniques. 8.9 Overlapped Slicing The problem was broken down into sub problems, which were reused several times. Overlapped slicing would duplicate an attribute with more than one column. Each column results in more attribute correlations [19].Horizontal and vertical partition was done by duplicating the attributes in more than one columns. Tuples were grouped together in horizontal partitioning. Attributes were correlated in vertical partitioning. Sensitive attribute would be placed in each column. Random permutation would take place on Sensitive Attribute [SA] column [19].The authors tried to provide protection against membership disclosure by distinguishing the original tuples from the fake tuples where the disclosure risk has been occurred. The matching stratergy in the overlapped slicing table is low. Overlapped slicing enhanced with slicing provided the membership disclosure protection. But, it leads to more attribute correlations and there would be a secrecy loss of privacy in some extent.
- 7. International Journal on Cybernetics & Informatics ( IJCI) Vol. 3, No. 1, February 2014 7 9. COMPARATIVE STUDY Techniques Dataset Parameter used Advantages Disadvantages K-Anonymity Market Basket Dataset Number of data points, Dimensionality of data space High correlation among the tuples More Number of dimensions would be violated ℓ-Diversity Adult Database Identifiers, Quasi- identifiers, Sensiti ve attribute Sensitive attribute would have at most same frequency Homogeneity and background knowledge attack has lacked t-closeness Pension scheme dataset Identifiers, Quasi- identifiers, Sensiti ve attribute Measure the distance between two probabilistic distribution that were indistinguishable from one another Information gain was unclear Km Anonymity Market Basket Dataset Distinct items, Maximum transaction size and Average transaction size on distinct items Similar evaluated approach on k items Loss of utility WFC Iris, Wine , Zoo Datasets Single, Complete and Average link Partition the records into equivalence classes Utility was still not achieved. Distributed K -Anonymity framework (DKA) Employee Dataset Public-key, Secret-key, Encryption Global Anonymization to ensure privacy Utility and potential were misused R-U Confidentiality Map Click Stream data Maximum transaction size, Average transaction size Maintain trade-off between privacy and utility Vulnerable to homogeneity attack Slicing Health care Dataset Identifier, Quasi- Identifier, Sensitiv e Attribute Randomization on sensitive attribute Utility and risk measures not matched Overlapped Slicing Health care Dataset Identifier, Quasi- Identifier, Sensitiv e Attribute Duplicate an attribute in more than one columns Utility was not achieved 10. CONCLUSION This paper, discussed about various anonymization techniques to preserve the privacy in data mining. Due to the large collection of information, it is important to maintain the Privacy. Number of anonymization techniques has been designed, still it remains an open problem on how to use the anonymization techniques effectively. The extra rows can be introduced to increase the redundancy in order to maintain the privacy. The more effective anonymization technique will be found to preserve the privacy and also ensure the privacy by cross-disciplinary fields.
- 8. International Journal on Cybernetics & Informatics ( IJCI) Vol. 3, No. 1, February 2014 8 REFERENCES [1] Charu C. Aggarwal, (2005), ‘‘On k-Anonymity and the Curse of Dimensionality”, Proceedings of the 31st VLDB Conference, Trondheim, Norway, pp.901-909 [2] Ashwin Machanavajjhala , Daniel Kifer,Johannes Gehrke, Muthuramakrishnan Venkita Subramanian, (2006),“ ℓ-Diversity : Privacy Beyond K-Anonymity”, Proc.International conference on Data Engineering.(ICDE),pp.24. [3] Anil Prakash, Ravindar Mogili ,(2012),‘‘Privacy Preservation Measure using t-closeness with combined l-diversity and k-anonymity”, International Journal of Advanced Research in Computer Science and Electronics Engineering (IJARC SEE)Volume 1, Issue 8,pp:28-33 [4] Yeye He, Jeffery Naughton .F, (2009), “Anonymization of Set Valued Data via Top Down Local Generalization”, Proc. International Conference on Very Large Databases (VLDB), pp.934- 945. [5] Wei Jiang , Chris Clifton, (2006)‘‘ A secure distributed framework for achieving k- anonymity”, the VLDB Journal , Vol.15, No.4, pp.316-333 [6] Chuang-Cheng Chiu , ChiehYuan Tsai, (2007),“ A k Anonymity Clustering method for Effective Data Privacy Preservation”, Springer journal on Verlag Berlin Heidelberg , pp.88-99. [7] Grigorios Loukides, Aris Gkoulalas - Divanis, and Jianhua Shao, (2012), ‘‘Assessing Disclosure Risk and Data Utility Trade-off in Transaction Data Anonymization”, International Journal of Software and Informatics, Vol.6, No. 3, pp.359-361 [8] Ravindra S, Wanjari Prof .Devi,(2013), “Improving the implementation of new approach for Data Privacy Preserving in Data Mining using slicing”. International Journal of Modern Engineering Research (IJMER), Vol. 3, Issue. 3. [9] L. Sweeney, (2002) “k-anonymity: a model for protecting privacy”, International Journal on Uncertainty, Fuzziness and Knowledge based Systems, pp. 557-570. [10] Yan Zhao, Ming Du, Jiajin Le, Yongcheng Luo,(2009), “ A Survey on Privacy Preserving Approaches in Data Publishing” in the First International Workshop on Database Technology and Applications [11] Mohnish Patel, Prashant Richariya, Anurag Shrivastava, (2013),‘‘A review paper on Privacy- Preserving Data Mining”, Review article on Scholars Journal of Engineering and Technology (SJET) , pp.359-361 [12] Agarwa, Srikan R., (2000) ‘‘Privacy Preserving Data Mining”, In Proc. ACM SIGMO, conference on management of data (SIGMOD’00), Dallas, TX,pp.439-450. [13] Benjamin c.m, Fung, ke wang, rui chen, philips s.yu ,(2010),‘‘Privacy Preserving Data Publishing:A Survey of Recent Development ”ACM Computing surveys, Vol.42, No.4, pp.523-553 [14] Byun, J.W, Kamra, A, Bertino, Li, N, (2007), ‘‘Efficient k-Anonymization Using clustering Techniques”. International Conference on Database Systems for Advanced Applications , pp.188-200. [15] N. Li, T. Li, and S. VenkataSubramanian, (2007), “ t-closeness : Privacy beyond k-anonymity and L -diversity,” Proc. International Conference on Data Engineering (ICDE), pp.106-115. [16] Sweeney L, (1996), “Replacing Personally Identifiable Information in Medical Records, the Scrub System”. Journal of the American Medical Informatics Association. [17] Charu C.Aggarwal, “A General survey of privacy preserving Data Mining Models and Algorithms”, IBM,T. J. Watson Research Centre [18] Tiancheng Li , Jian Zhang , Ian Molloy ,(2012),“Slicing: A New Approach for Privacy Preserving Data Publishing” IEEE Transaction on KDD. [19] B.Vani, D.Jayanthi, (2013), “Efficient Approach for Privacy Preserving Microdata Publishing Using Slicing” IJRCTT.