Process mining relies on high-quality event logs for accurate analysis and decision-making. However, real-life event logs suffer from various types of data quality issues. Existing solutions are retroactive, detecting and repairing data quality issues after their occurrence in event logs. A more permanent and cost-effective solution would be to prevent these issues proactively, “prevention is better than the cure”. This paper proposes a data-driven approach (eRooMiner) for identifying root causes of typical data quality issues in event logs to facilitate their prevention. The eRooMiner approach builds upon the typical data quality issues identified in the collection of eleven event log imperfection patterns, with a focus on imperfect labels which have the same meaning but different syntax. By learning from the data recorded in event logs, eRooMiner bridges the gap between theoretical and data-driven root cause analysis of event log imperfection patterns. The approach utilises machine learning and AI techniques to estimate the most probable root cause of specific data imperfections. The approach has been implemented and evaluated using real-life event logs and stakeholders. The results show that the proposed approach can correctly detect the potential root cause(s) of imperfect labels in various scenarios.