Differential Privacy Protection for Support Vector Machines for Nonlinear Classification

Currently, private data leakage and nonlinear classification are two challenges encountered in big mining. In particular, few studies focus on these issues support vector machines (SVMs). this paper, to effectively solve them, we propose a novel framework based the concepts of differential privacy (DP) kernel functions. This can allocate budgets add artificial noise different SVM locations simultaneously, which makes perturbation process freer more delicate. addition, under framework, three algorithms, DP SVMs that perturb training set, function, utilize mixed (DPSVM-TDP, DPSVM-KFP, DPSVM-MP, respectively), all realize accurate while ensuring users’ is not violated. Moreover, conduct analysis algorithms prove they satisfy ε , 0 − DP. Finally, experiments evaluate terms aspects compare them with DPSVM dual-variable (DVP) algorithm (DPSVM-DVP) determine optimal method. The results show DPSVM-KFP achieve highest utility strictest protection shortest running time.