Predict state of newly arriving process instance (case)



Mahsa Pourbafrani

Wissenschaftliche Mitarbeiterin


+49 241 80 21906



MSc Thesis Project

Title: Predict state of newly arriving process instance (case)

Author: Shreya Kar

Supervisor: Mahsa Pourbafrani

1st examiner: Prof.Dr. ir. Wil M.P. van der Aalst

2nd examiner: Prof. Dr. Stefan Decker


Predictive process monitoring is an application area of process mining that predicts the future state, e.g., outcome or time to completion, for an ongoing process instance. The timely identification of delay or deviation in process behavior for an ongoing case allows organizations to take preventive actions. There are many predictive process monitoring approaches. However, most approaches learn to make predictions based on the assumption that cases are isolated, leading to poor prediction accuracy. In reality, most processes contain inter-case dynamics because of cases sharing resources. We propose an inter-case dynamics aware predictive process monitoring approach for remaining time prediction to address this problem. First, we present a set of techniques to analyze inter-case dynamics within a subset of process segments that cause high prediction errors. Then we utilize the insights gained to create features for inter-case dynamics caused by batching and non-batching patterns. Our proposed approach tries to improve the approach presented by Klijn et al. [1] by considering a more comprehensive range of inter-case patterns while making predictions. We also try to improve the process presented in the previous approach to detect segments subject to inter-case dynamics by introducing some automation. We evaluate our proposed remaining time prediction approach with two publicly available real-life event logs. The results show that the inclusion of the created inter-case features in remaining time predictions leads to more accurate and interpretable predictions.