The main purpose of a workflow management system is to support and to 
automate the execution of business processes. One important aspect of 
automation is the integration of time and time restrictions (e.g. check 
the deadline of a task). The violation of temporal constraints forces 
special-actions to handle the escalated situation, which is likely to 
increase the cost of the business process.
A workflow system which is enabled to predict such temporal problems 
opens up new vistas on handling escalations: it is possible to intervene 
long bevore the violation normally would be spotted, thus saving both time 
and resources. One major problem of these predictive workflow management 
systems is the impossibility to forecast the exact duration of a business 
process or its activities, since different instances of one business process 
will have varying execution-durations. The approach illustrated in this master 
thesis solves the above mentioned problem. The workflow log holds invaluable 
information about past workflow executions: It is possible to extract probability 
distributions from the log and map them on activity durations, branching probabilities 
and the number of iteration loops of a workflow process definition. These are the 
prerequisites to calculate a timed workflow graph, which holds a probability 
distribution of the estimated remaining process execution time für each workflow 
activity. This graph provides the opportunity to forecast the certainty for the 
process to terminate without violating temporal constraints or to terminate within 
a given period of time. In addition it is explained how to use the timed graph 
to evaluate temporal constraints during the build-time of a workflow process. 
Furthermore some scenarios are presented which point out the possibilities to 
use the graph as a decision-supporting mechanism for the workflow engine during run-time.