• crowding
• data-mining
• Emergency Department
• forecasting
• machine learning
• waiting time

This thesis aims at providing a robust predictive model that accurately estimates the waiting time of patients in Emergency Department (ED). This objective is achieved by developing a methodology that integrates the Process-Mining approach with the Data-Mining approach.
Process-Mining exploits ED event logs to derive information about patient flow and the congestion state of the system. This information is transformed into valuable predictors that feed Machine Learning algorithms. Several learning algorithms are compared, such as Regularized Linear Regression, Random Forest, Support Vector Regression, Neural Network and Ensemble Method. The developed methodology is applied to a real case, an ED in Tuscany. The best performing predictive model is selected and a refinement is conducted to select the most powerful predictors.
Finally, this thesis highlights the managerial implications of predicting waiting times for both patients and ED managers. Providing patients with accurate waiting time information positively affect their behavior, increasing their tolerance to waiting. This leads to greater patient satisfaction and reduces the number of patients who decide to leave the ED without being seen by a physician. From a hospital perspective, predicting the waiting time allows the hospital managers to be constantly informed of the crowded state of ED and thus supports them in managing resources efficiently.