Seminar: Privacy-Preserving Feature Selection

Javad Rahimipour Anaraki
Ph.D. Oral Comprehensive
Supervisors: Drs. Saeed Samet & Wolfgang Banzhaf

Privacy-Preserving Feature Selection

Department of Computer Science
Thursday, February 9, 2017, 2:00p.m., Room EN 2022


In machine learning and pattern recognition, feature selection is the process of selecting the most instructive features of a problem while removing unnecessary ones. This process plays an important role by reducing the dimension of datasets in health data (e.g., functional near-infrared spectroscopy neural signals, internal jugular vein data processing) and in bioinformatics (e.g., gene expression, molecular interaction networks). One of the major drawbacks of trending feature selection methods is a lack of privacy-preserving safeguards that make them impractical in sensitive data processing tasks, such as in the health sector. This research will investigate securing four existing feature selection algorithms, propose a new feature selection method based on a system of equations and also introduce a feature selector-classifier using linear genetic programming that prioritizes privacy and security. All proposed methods will be integrated into a machine-learning package called Lucas specifically designed and enhanced for privacy-preserving machine-learning algorithms.