E N D
Comparative Evaluation of Four Different Sensitive Tabular Data Protection Methods Using a Real Life Table Structure of Complex Hierarchies and LinksPopulated with Artificial DataRamesh A. DandekarEnergy Information AdministrationWashington DC(Ramesh.Dandekar@EIA.DOE.GOV) UNECE2007 – 17-19 December 2007 Manchester, United Kingdom
Tabular Data Protection Methods* • Classical LP Based Cell Suppression • Network Flow Based Cell Suppression (USBC) • LP Based Synthetic Tabular Data / CTA (Dandekar 2001) • Micro Data Level Noise Addition (USBC) P = 10 % Rule Used * Uses Proprietary Research Tools Manchester, United Kingdom
Two Three Dimensional HYPOTHETICALTablesLinked in Four Dimensional Space1st Table: “Volumes by Grade, Sales Type, PAD District, and State” , and2nd Table: “Volumes by Formulation, Sales Type, PAD District, and State” Manchester, United Kingdom
Regular Midgrade Premium Total All Grades Conventional Oxygenated Reformulated Total All Formulations 1st Table 2nd Table GradesFormulations Manchester, United Kingdom
Grades MISSING PORTION Formulations 2nd Table By Formulations 1st Table By Grades Total Four Layers: 1) DTW 2) Rack 3) Bulk 4) Total Corresponding to each PAD, State and US Level Cell Manchester, United Kingdom
1,000 Synthetic Micro Data Records Containing Six Variables Four Categorical Variables • 51 States • 3 Grade Types • 3 Sale Types • 3 Formulation Types One Magnitude Variable One Sample Weight Variable Manchester, United Kingdom
Classical Cell Suppression 294 Suppressions Sensitive Cells Fully Protected Network Flow Method 479 Suppressions Sensitive Cells Fully Protected 3 exact Disclosures of non-sensitive cells Comparative Evaluation of Cell Suppression Methods Manchester, United Kingdom
CTA vs NOISE - TABULAR DATA QUALITY 633 1432 Manchester, United Kingdom
THANK YOU!ADDITIONAL INFORMATION FROMhttp://mysite.verizon.net/vze7w8vk/ Manchester, United Kingdom