Predicting Business Process Events in Presence of Anomalous IT Events

Predictive process monitoring is concerned with predicting business process behaviour, such as the next activity in a case, the time for completion of an activity, and the remaining execution trace of a case. The primary goal is to predict a violation or a delay before its occurrence and take appropriate corrective measures proactively. Existing literature largely focuses on predictions considering primarily the knowledge of the process context such as the previous sequence of activities of the executing trace or the human resources performing the task. The recent move towards migrating business processes in the cloud has resulted in multi-cloud configurations with parts of the process run by different cloud providers and parts run on premises. This has made the execution of business processes heterogeneous, distributed, and complex. In these conditions, the execution of the process requires the additional context of the underlying IT systems executing the process. Sometimes anomalous IT events can adversely affect certain parts of a process. Diagnosing the impact of such anomalous IT events is exacerbated in these complex deployments. In this paper we study whether existing process prediction frameworks can deal with such anomalous IT events. We propose an approach that considers IT events as input to identify the cases that are likely to be impacted by such anomalous events, particularly the delay caused by such events. Our approach performs competitively compared to baselines when predicting the next event in a case and the completion time of each event.