Deep Learning for Predictive Business Process Monitoring: Review and Benchmark (Summary)

Resumen:

Predictive monitoring is a discipline that aims to predict how an ongoing business process case will unfold using the information from a business process event log. Recently, deep neural networks have gained much attention for this task due to their good results. Unfortunately, the high number of neural network architectures, the different ways of encoding the partial traces and events, the number of predictive tasks available, and, sometimes, the difficulty of quantifying the differences and contributions between the works may complicate the task of defining what has already been done, what can be researched and future research directions. Furthermore, and more importantly, approaches are applied to a reduced number of datasets with significantly different experimental setups. This makes it difficult for new researchers to compare their new approach with existent state-of-the-art works. Thus, this paper identifies the most relevant deep learning approaches for predictive monitoring. Then, using these approaches as a starting point, we select those with their implementation available and perform exhaustive experimentation of 10 different approaches and a statistical comparison in a fair setting using 12 publicly available event logs. Then, we use the results to highlight the most relevant differences between them.

Palabras Clave:

Business Process Monitoring - Deep Learning - Neural Networks - Process Mining - Systematic literature review

Autor(es):

• Efrén Rama Maneiro {efren.rama.maneiro@usc.es}
  Universidad de Santiago de Compostela - Spain
  
• Juan C. Vidal {juan.vidal@usc.es}
  University of Santiago de Compostela - Spain
  
• Manuel Lama Penin {manuel.lama@usc.es}
  University of Santiago de Compostela - Spain
  

Handle:

11705/JCIS/2022/013

Descargas:

Acceso a los detalles haciendo click aquí.