| For the past decade or so, the needs for organizations to share data for knowledge discovery and data mining have increased significantly. Meanwhile, privacy issues have become widely recognized and motivated the research on privacy-preserving data mining (PPDM). In this dissertation, we study frameworks, models, methodology and evaluations of two approaches of PPDM: the privacy-preserving data publishing and the privacy-preserving knowledge model sharing.;In privacy-preserving data publishing, data owners release anonymized versions of their data. We propose new privacy measures, utility specification, and anonymization algorithms based on data generalization techniques and consider both one-time and multi-time data publishing in an environment consisting of a single data source.;In privacy-preserving knowledge model sharing, data owners release knowledge models learned from their data. We propose new privacy measures and algorithms for data owners to published privacy-preserving decision trees learned from local data in an environment of multiple data sources and for knowledge users to learn global knowledge models from local published knowledge models. |
