Modelling a secure support vector machine classifier for private data

M. Sumana, K. S. Hareesha

Research output: Contribution to journalArticle

Abstract

Privacy preserving data mining engrosses in drawing out information from distributed data without disclosing sensitive information to collaborating sites. This paper aims on the construction of a vertically distributed privacy preserving support vector machine classifier. The learning model is build for datasets, where one of the collaborating parties comprises the dependent attribute. Furthermore, the amount of privacy, computation speed and the accuracy of our classifier outperform other benchmark algorithms. Privacy of the perceptive attributes values of the cooperating sites are retained while performing secure computations. Collaborative classification is performed using these attributes. The site with the dependent attribute is the master site that initiates the process of secure computation to identify support vectors. Homomorphic property is used to protectively compute the data matrix on records/tuples available at sites. The recommended nonlinear privacy preserving classifier provides an accuracy equivalent to the non-privacy undistributed SVM classifier which uses all the attributes directly.

Original languageEnglish
Pages (from-to)25-40
Number of pages16
JournalInternational Journal of Information and Computer Security
Volume10
Issue number1
DOIs
Publication statusPublished - 01-01-2018

Fingerprint

Support vector machines
Classifiers
Data mining

All Science Journal Classification (ASJC) codes

  • Software
  • Safety, Risk, Reliability and Quality
  • Hardware and Architecture
  • Computer Networks and Communications

Cite this

@article{e5c08f5967c9420e9edc33327c32c400,
title = "Modelling a secure support vector machine classifier for private data",
abstract = "Privacy preserving data mining engrosses in drawing out information from distributed data without disclosing sensitive information to collaborating sites. This paper aims on the construction of a vertically distributed privacy preserving support vector machine classifier. The learning model is build for datasets, where one of the collaborating parties comprises the dependent attribute. Furthermore, the amount of privacy, computation speed and the accuracy of our classifier outperform other benchmark algorithms. Privacy of the perceptive attributes values of the cooperating sites are retained while performing secure computations. Collaborative classification is performed using these attributes. The site with the dependent attribute is the master site that initiates the process of secure computation to identify support vectors. Homomorphic property is used to protectively compute the data matrix on records/tuples available at sites. The recommended nonlinear privacy preserving classifier provides an accuracy equivalent to the non-privacy undistributed SVM classifier which uses all the attributes directly.",
author = "M. Sumana and Hareesha, {K. S.}",
year = "2018",
month = "1",
day = "1",
doi = "10.1504/IJICS.2018.089587",
language = "English",
volume = "10",
pages = "25--40",
journal = "International Journal of Information and Computer Security",
issn = "1744-1765",
publisher = "Inderscience Enterprises Ltd",
number = "1",

}

Modelling a secure support vector machine classifier for private data. / Sumana, M.; Hareesha, K. S.

In: International Journal of Information and Computer Security, Vol. 10, No. 1, 01.01.2018, p. 25-40.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Modelling a secure support vector machine classifier for private data

AU - Sumana, M.

AU - Hareesha, K. S.

PY - 2018/1/1

Y1 - 2018/1/1

N2 - Privacy preserving data mining engrosses in drawing out information from distributed data without disclosing sensitive information to collaborating sites. This paper aims on the construction of a vertically distributed privacy preserving support vector machine classifier. The learning model is build for datasets, where one of the collaborating parties comprises the dependent attribute. Furthermore, the amount of privacy, computation speed and the accuracy of our classifier outperform other benchmark algorithms. Privacy of the perceptive attributes values of the cooperating sites are retained while performing secure computations. Collaborative classification is performed using these attributes. The site with the dependent attribute is the master site that initiates the process of secure computation to identify support vectors. Homomorphic property is used to protectively compute the data matrix on records/tuples available at sites. The recommended nonlinear privacy preserving classifier provides an accuracy equivalent to the non-privacy undistributed SVM classifier which uses all the attributes directly.

AB - Privacy preserving data mining engrosses in drawing out information from distributed data without disclosing sensitive information to collaborating sites. This paper aims on the construction of a vertically distributed privacy preserving support vector machine classifier. The learning model is build for datasets, where one of the collaborating parties comprises the dependent attribute. Furthermore, the amount of privacy, computation speed and the accuracy of our classifier outperform other benchmark algorithms. Privacy of the perceptive attributes values of the cooperating sites are retained while performing secure computations. Collaborative classification is performed using these attributes. The site with the dependent attribute is the master site that initiates the process of secure computation to identify support vectors. Homomorphic property is used to protectively compute the data matrix on records/tuples available at sites. The recommended nonlinear privacy preserving classifier provides an accuracy equivalent to the non-privacy undistributed SVM classifier which uses all the attributes directly.

UR - http://www.scopus.com/inward/record.url?scp=85041429535&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85041429535&partnerID=8YFLogxK

U2 - 10.1504/IJICS.2018.089587

DO - 10.1504/IJICS.2018.089587

M3 - Article

AN - SCOPUS:85041429535

VL - 10

SP - 25

EP - 40

JO - International Journal of Information and Computer Security

JF - International Journal of Information and Computer Security

SN - 1744-1765

IS - 1

ER -