Building anonymised database samples

Dissertação para obtenção do Grau de Mestre em Engenharia Informática

In this work we propose Anonym Database Sampler (ADS), a flexible and modular system capable of extracting an anonymised, consistent and representative sample from a relational database. ADS was envisioned for use in testing and development environments. To this end, a sample specification input is requested from the user, that is used by ADS’s sampling engine to perform a stratified random sample. Afterwards a First-choice hill climbing algorithm is applied to the sample, optimising the selected data towards the specified requisites. Finally, if some restrictions are still to be met, tuples and/or keys modifications are performed, ensuring that the final sample fully complies with the initial sample specification. While having a representative and sound database that developers can use in these environments can be a great advantage, we assume that this representativeness does not need to comply with a truly statistical representativity, which would be much harder to obtain. Thereby, ADS samples are not appropriate for any kind of statistical data analysis. After the sample being successfully extracted, due to the sensitivity of the data contained in most organisation databases, a data anonymisation step is performed. The sampled data is consistently enciphered and masked, preventing data privacy breaches that could occur by delivering to developers a database containing some real operational data.