Differential Privacy Principal Component Analysis for Support Vector Machines

Algorithm 1

Differential privacy principal component analysis-support vector machine (DPPCA-SVM).

	Input: data , samples n, attributes d, privacy budget ;
	Output: classification decision function
(1)	Compute covariance matrix of input data ;
(2)	Compute symmetric noise matrix E, each element in E is sampled from Laplace distribution with a scaling of ;
(3)	Add noise to the original covariance matrix ;
(4)	Compute eigenvalues and corresponding eigenvectors of the noise covariance matrix ;
(5)	Select first k eigenvectors to determine the low-dimensional data ;
(6)	Compute classification function ;