Techniques disclosed herein relate to removing potentially identifying features of a specific subject from a data set to prevent re-identification of the subject using an external data source. In various embodiments, the data set contains, as potential identifying features of the specific subject, multiple bursts of temporally-proximate events. Time blocks within the data set can be identified to capture one or more of the bursts of temporally-proximate events for the specific subject. Adding random time shifts for each time block can add noise to the data set and remove or obfuscate the identifying features of a specific subject to generate a time shifted data set.