Understanding business process dynamics through system-level process mining

Process mining studies how organizational processes are executed. It analyzes event logs recorded by information systems to reveal how processes actually run, detect inefficiencies, and support improvement. Most existing methods examine process instances and events in isolation and assume that processes remain stable over time. In reality, business processes are dynamic systems whose overall behavior changes as operating conditions evolve. This dynamic behavior is often ignored, which limits the ability to address many process mining problems. This thesis shows how modeling business process dynamics can advance process mining through system-level analysis. It focuses on high-level process properties that emerge from many concurrent instances and offers four contributions: a taxonomy and framework for concept drift, a computer vision method for drift detection, a framework for steady-state detection, and a data-driven method for assessing process resilience. The approaches are evaluated on synthetic and real data using established algorithm-engineering practices.