A Privacy-Preserving Twin Support Vector Machine Classifier for Vertical Partitioned Data

摘要

In this paper, a novel privacy-preserving binary classifier termed as break Privacy Preservation Twin Support Vector Machine (PPTWS VM) has been proposed. The PPTWSVM formulation is motivated by the Privacy-Preserving Support Vector Machine (PPSVM) formulations of Mangasarian and Wild (Mangasarian et al. in ACM Trans Knowl Discov Data 2(3),12, 2008 [1]; Mangasarian and Edward in Privacy-preserving classification of horizontally partitioned data via random Kernels, 2008 [2]; Mangasarian and Edward in Privacy-preserving random Kernel classification of checkerboard partitioned data. Data mining. Springer, USA, 2010 [3]) and Twin Support Vector Machine (TWSVM) formulation of Jayadeva et al. (IEEE Trans Pattern Anal Mach Intell 29(5):905-910, 2007 [4]; Khemchandani and Chandra in Twin support vector machines: models, extensions and applications. Springer, 2016 [5]). Similar to PPSVM, PPTWSVM also employs the random kernel technique for preserving the privacy of participating entities which are holding the different feature columns of the representing data. An extensive numerical implementation on UCI benchmark datasets confirms that PPTWSVM is faster than PPSVM in the training phase and owns better generalization ability.