Are Sequential Patterns Shareable? Ensuring Individuals’ Privacy

Individuals’ actions like smartphone usage, internet shopping, bank card transaction, watched movies can all be represented in form of sequences. Accordingly, these sequences have meaningful frequent temporal patterns that scientist and companies study to understand different phenomena and business processes. Therefore, we tend to believe that patterns are de-identified from individuals’ identity and safe to share for studies. Nevertheless, we show, through unicity tests, that the combination of different patterns could act as a quasi-identifier causing a privacy breach, revealing private patterns. To solve this problem, we propose to use \(\epsilon \)-differential privacy over the extracted patterns to add uncertainty to the association between the individuals and their true patterns. Our results show that its possible to reduce significantly the privacy risk conserving data utility.