Analyzing and repairing overlapping work items in process logs

Context: In real life logs, it often happens that some human resources appear to have more than one task active concurrently, thus resulting in human multitasking. However, tasks that require some intellectual effort cannot be executed in parallel in real life. This misalignment between what actually happens and what is registered in the logs, however, is not reflected in the output of the different log-based performance measuring approaches, thus compromising the quality of the computed metrics. Objective: We introduce a novel approach to rewrite events in process execution logs for multitasking human resources. The approach is based on two typical human work patterns, the queuing and stacking patterns. The rewrite aims at serializing multi tasks for the same resource based on the work pattern detected. Thus, possibly better performance measures can be obtained. Method: We defined a quantitative approach to detect multitasking human performers and resolve them by serialization. The approach is prototyped and evaluated on a set of real-life software development process logs. Results: Our results show that the proposed approach contributes to find better results when log-based performance analysis techniques are applied to the repaired logs in comparison to the original logs. Conclusions: The work shows that based on the human work patterns, stacking or queuing, logs can be enhanced, so as to be possibly closer to what happened in the reality and to allow for more accurate performance measures.