Optimal event log sanitization for privacy-preserving process mining

Abstract

Event logs that originate from information systems enable comprehensive analysis of business processes. These logs serve as the starting point for the discovery of process models or the analysis of conformance of a log with a given specification. However, logs potentially contain personal information about individuals involved in process execution. In this paper, we therefore address the risk of privacy attacks on event logs. Specifically, we rely on group-based privacy guarantees instead of noise insertion in order to enable anonymization without adding new behaviour to the log. To this end, we propose two new algorithms for event log sanitization that provide privacy guarantees in terms of k-anonymity for the behavioural perspective of a process and t-closeness for sensitive information associated with events. The algorithms thereby avoid the disclosure of employee identities, prevent the identification of employee membership in the log, and preclude the characterization of employees based on sensitive attributes. Our algorithms overcome the limitations of an existing, greedy algorithm, providing users with a trade-off between computational complexity and the utility of the sanitized event log for downstream analysis. Our Experiments demonstrate that sanitization with our algorithms generates event logs of higher utility compared to the state of the art.