What is it about?
Organizations record the execution of their business processes as sequences of events. Predictive process monitoring uses these historical records to anticipate what is likely to happen next in a process that is still running. This paper introduces PROPHET, an artificial intelligence approach that predicts the next activity of an ongoing business process while also explaining its prediction. PROPHET represents process executions as heterogeneous graphs. Different types of nodes describe events and their characteristics, while different types of connections capture the relationships among events and their attributes. A graph neural network then learns from this structured representation to predict the most likely next activity. The approach also identifies which events, characteristics and relationships contributed most strongly to each prediction.
Featured Image
Photo by GuerrillaBuzz on Unsplash
Why is it important?
Accurate predictions can help organizations anticipate delays, errors and undesirable outcomes. However, predictions generated by complex deep-learning models are often difficult for people to understand. This lack of transparency can limit their use in operational and high-stakes settings. Process managers need to know not only what the model predicts, but also why it has reached that conclusion. PROPHET addresses this challenge by combining predictive accuracy with explainability. Its explanations can help users understand the evidence behind a prediction, evaluate whether it is reasonable and make more informed decisions about an ongoing process.
Perspectives
Our objective was to move predictive process monitoring beyond black-box predictions. Business processes contain multiple relationships among activities, event attributes and execution contexts. Representing this information as a heterogeneous graph allows the predictive model to capture these relationships explicitly. At the same time, the explainability component reveals which parts of the process execution influenced the predicted next activity. We believe this combination of prediction and explanation is an important step towards more transparent and trustworthy process-aware artificial intelligence. Key takeaways - PROPHET predicts the next activity of an ongoing business process. - Process traces are represented as heterogeneous graphs. - Different node and connection types capture events, their characteristics and their relationships. - Graph Attention Networks learn which information is most relevant for prediction. - A modified GNNExplainer identifies the events, characteristics and relationships that influenced each result. - The approach combines predictive performance with interpretable, case-specific explanations.
Prof. Donato Malerba
Universita degli Studi di Bari Aldo Moro
Read the Original
This page is a summary of: PROPHET: Explainable Predictive Process Monitoring With Heterogeneous Graph Neural Networks, IEEE Transactions on Services Computing, November 2024, Institute of Electrical & Electronics Engineers (IEEE),
DOI: 10.1109/tsc.2024.3463487.
You can read the full text:
Contributors
The following have contributed to this page
