Conclusion

In this paper, we propose an approach for analyzing event logs to identify improvement opportunities in processes with batch processing activities. In addressing RQ1, we define three types of waiting times associated with batch processing, namely waiting time for batch accumulation and waiting time of a ready batch in parallel, sequential, and concurrent batch processing, and waiting time for other cases to be processed in sequential and concurrent batch processing. We use these definitions to identify waiting times from event logs.

In our approach, we first discover different types of batch processing from an event log. For this, we extend existing research by also considering the waiting times associated with batch processing and discovering their impact on the CTE of batch processing activities (RQ2). Finally, we identify batch processing inefficiencies by measuring the impact of batch processing waiting times on the activity CTE. This enables analysts to identify where there are improvement opportunities and where to target the process changes (RQ3). We evaluated the approach with synthetic data and a real-life event log. The evaluation indicates that our approach can provide analysts with insights on potential batch processing inefficiencies. Thus, the analysts can take a data-driven approach to improve batch processing efficiency.

As a result of the experimentation, we detected potential improvements in the approach. Our approach uses all observations available per batch for the discovery of batch activation rules and reports on confidence and support. This process can be improved by establishing training/test partitions to discover and validate the rules. The approach applicability is, however, limited to the event logs that have data on the resources, start and end timestamps. If the log miss any of these data, the approach cannot be executed and the analyst is informed accordingly. In addition, the non-working periods of resources are currently part of the measured waiting times. Our approach could be extended by adding calendar information to improve the accuracy of both measures. Finally, the evaluation could be extended by including other synthetic event logs and a larger real-life event log with multitasking.

As future work, we plan to implement a what-if simulation analysis for batch processing activities to identify the impact of particular changes on the CTE. This would allow analysts to change the parameters of batch processing activities and explore what changes would improve batch processing performance and by how much.