Predicting Waiting and Service Times in Emergency Departments through Machine Learning and Process Mining

Emergency Departments (EDs) have gained particular attention under the pres-sure of the public opinion and national authorities. The EDs’ performance has in fact the highest impact on patient care and is highly relevant in public opinion, e.g. public debate on excessive waiting time or misdiagnosis. To better handle the ED resources, which tend to be particularly limited, and improve the ED perfor-mances, it would be important for health managers to understand what the current situation is and its potential evolution. This may allow to prevent overcrowding and/or to react dynamically to likely-critical conditions, anticipating situations in which waiting and service times become unacceptable and practitioners are sub-jected to excessive pressure. However, the complex nature of EDs imposes nu-merous challenges on the prediction of ED overcrowding and of waiting/service time trends. Insert in this context, this research aims to develop a forecasting system to pre-dict the waiting and service times for the radiological unit of an ED, using real data from an Italian hospital. The system monitors the current state of the ED processes, also exploiting Process Mining, and makes waiting and service time predictions through Machine Learning techniques. The results may have significant practical implications for ED and hospitals. In-deed, the developed system allows real-time monitoring of the radiological unit and warns of deteriorating performance before they happen. This information would be essential for ED managers to make more responsive and proactive ac-tions if expected long waiting times can be brought forward.