Activity failure prediction based on process mining

Based on the state of the art of process mining, we can conclude that quality characteristics (failure rate metrics or loops) are poorly represented or absent in most predictive models that can be found in the literature. The main goal of this present research work is to analyze how to learn prediction model defining failure as response variable. A model of this type can be used for active real-time-controlling (e. g. through the reassignment of workflow activities based on prediction results) or for the automated support of redesign (i.e., prediction results are transformed in software requirements used to implement process improvements). The proposed methodology is based on the application of a data mining process because the objective of this work can be considered as a data mining goal.