On supporting K-anonymisation and L-diversity of crime databases with genetic algorithms in a resource constrained environment
Master Thesis
2015
Permanent link to this Item
Authors
Supervisors
Journal Title
Link to Journal
Journal ISSN
Volume Title
Publisher
Publisher
University of Cape Town
Department
Faculty
License
Series
Abstract
The social benefits derived from analysing crime data need to be weighed against issues relating to privacy loss. To facilitate such analysis of crime data Burke and Kayem [7] proposed a framework (MCRF) to enable mobile crime reporting in a developing country. Here crimes are reported via mobile phones and stored in a database owned by a law enforcement agency. The expertise required to perform analysis on the crime data is however unlikely to be available within the law enforcement agency. Burke and Kayem [7] proposed anonymising the data(using manual input parameters) at the law enforcement agency before sending it to a third party for analysis. Whilst analysis of the crime data requires expertise, adequate skill to appropriately anonymise the data is also required. What is lacking in the original MCRF is therefore an automated scheme for the law enforcement agency to adequately anonymise the data before sending it to the third party. This should, however, be done whilst maximising information utility of the anonymised data from the perspective of the third party. In this thesis we introduce a crime severity scale to facilitate the automation of data anonymisation within the MCRF. We consider a modified loss metric to capture information loss incurred during the anonymisation process. This modified loss metric also gives third party users the flexibility to specify attributes of the anonymised data when requesting data from the law enforcement agency. We employ a genetic algorithm(GA) approach called "Crime Genes"(CG) to optimise utility of the anonymised data based on our modified loss metric whilst adhering to notions of privacy denned by k-anonymity and l-diversity. Our CG implementation is modular and can therefore be easily integrated with the original MCRF. We also show how our CG approach is designed to be suitable for implementation in a developing country where particular resource constraints exist.
Description
Keywords
Reference:
Verster, C. 2015. On supporting K-anonymisation and L-diversity of crime databases with genetic algorithms in a resource constrained environment. University of Cape Town.