Time series anonymization is an important problem. One prominent example of time series are energy consumption records, which might reveal details of the daily routine of a household. Existing privacy approaches for time series, e.g., from the field of trajectory anonymization, assume that every single value of a time series contains sensitive information and
reduce the data quality very much. In contrast, we consider time series where it is combinations of tuples that represent personal information. We propose (n; l; k)-anonymity, geared to anonymization of time-series data with minimal information loss, assuming that an adversary may learn a few data points. We propose several heuristics to obtain (n; l; k)-anonymity,
and we evaluate our approach both with synthetic and real data. Our experiments confirm that it is sufficient to modify time series only moderately in order to fulfill meaningful privacy requirements.
Pattern-sensitive Time-series Anonymization and its Application to Energy-Consumption Data
Journal: OJIS 2014, 1(1), Pages 3-22