“Secure ” Logistic Regression of Horizontally and Vertically Partitioned Distributed Databases

Privacy-preserving data mining (PPDM) techniques aim to construct efficient data mining algorithms while main-taining privacy. Statistical disclosure limitation (SDL) tech-niques aim to preserve confidentiality but in contrast to PPDM techniques also aim to provide access to statistical data needed for “full ” statistical analysis. We draw from both PPDM and SDL paradigms, and address the prob-lem of performing a “secure ” logistic regression on pooled data collected separately by several parties without directly combining their databases. We describe “secure ” Newton-Raphson protocol for binary logistic regression in the case of horizontally and vertically partitioned databases using secure-mulity party computation. 1