Reidentifying Anonymised Data Using Machine Learning
Year: 2020
Type: dissertation
Abstract: In a digital world dominated by data, where vast amounts of data is collected about individuals each day, privacy is becoming an ever growing concern. With the introduction of wearable fitness trackers and smartwatches and their growing popularity, potentially sensitive health-related data is now also being collected about individuals, which further emphasises the need for privacy. Common privacy-preserving actions today usually involves anonymisation of personal data, to prevent the person from being identifiable in a dataset, should it be sold, shared or otherwise published. Privacy regulations are in place, and in the process of improving, such as with the introduction of the EU General Data Protection Regulation (GDPR) being recently introduced in 2018. However, as we will see in this thesis, these efforts to protect privacy may not be sufficient. With deep learning technology blurring the lines between identifiable and non-identifiable personal data due to its tremendous learning capabilities and ability to recognise patterns and correlations between data, there are concerns that current data anonymisation processes are no longer sufficient to protect a person's privacy. To test this theory, we will apply deep learning to an anonymised Fitbit dataset in an attempt to reidentify the participants within the dataset. In this thesis, we will be implementing a Siamese neural network architecture as a solution to this problem.
Author: Martin Kristoffer Svensen
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W3180825805