Privacy Preserving PCA for Multiparty Modeling
share
In this paper, we present a general multiparty model-ing paradigm with Privacy Preserving Principal ComponentAnalysis (PPPCA) for horizontally partitioned data. PPPCAcan accomplish multi-party cooperative execution of PCA un-der the premise of keeping plaintext data locally. We also pro-pose implementations using two techniques, i.e., homomor-phic encryption and secret sharing. The output of PPPCA canbe sent directly to data consumer to build any machine learn-ing models. We conduct experiments on three UCI bench-mark datasets and a real-world fraud detection dataset. Re-sults show that the accuracy of the model built upon PPPCAis the same as the model with PCA that is built based on cen-tralized plaintext data.
READ FULL TEXT