Loading...

Predictive Business Process Monitoring Using Machine Learning Algorithms

Feiz, Roya | 2023

61 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56668 (01)
  4. University: Sharif University of Technology
  5. Department: Industrial Engineering
  6. Advisor(s): Hassannayebi, Erfan
  7. Abstract:
  8. In order to survive in today's business world, which is changing at a very fast pace, organizations can detect deviations even before they occur, quickly and with a high percentage of confidence, by analyzing their processes, in order to prevent disruptions in the processes. by monitoring the information systems that automatically execute business processes, it is possible to ensure the correct implementation of the existing processes. For this purpose, various techniques for monitoring business processes have been presented so that managers have a comprehensive and real view of how implement processes and be able to identify possible deviations in the future and try to fix them because the occurrence of some deviations can impose very heavy costs on the organization. In recent years, the field of predictive monitoring of business processes with the help of machine learning algorithms and In a more advanced way, deep learning has been expanded. In general, the goal of this field is to be able to predict the future behavior of a running process by using data from past events. Due to the nature of event logs, which is related to past data, the use of recurrent neural networks to predict the future behavior of a running process such as LSTM is common in this field. However, due to the fact that in practice the execution sequences of the processes are very long and these networks cannot store a lot of information in a vector with a fixed length and in a long steps, the accuracy of the resulting predictions will decrease. Which part of the process is given more importance, in this way, the past information will be better preserved. In this research, a sequence-to-sequence structure with the blocks of a recurrent neural network with attention mechanism was presented, which is unlike other suffix prediction researches that are only repetition of predicted next activities, by providing two sections of encoder and decoder, it made it possible to learn the representation of prefixes in the encoder phase and to predict the suffix of activities in another separate phase under the title of decoder, and in this way the connection between events is preserved. and is learned by the proposed structure. The proposed structure is able to simultaneously predict the extension of future activities and the remaining time to complete the path of a running item. In the inference phase of this structure, an innovative search algorithm is used to improve the selection of extension activities to was used. In order to evaluate the results, four real data sets and two basic articles were considered. The results show the superiority of the proposed models compared to the basic articles. The proposed structures have been able to significantly increase the accuracy of predictions. Also The proposed structures had less training time than the basic articles. Also, the accuracy of the predictions increased by applying the heuristic search algorithm compared to other sequence activity selection techniques
  9. Keywords:
  10. Process Mining ; Predictive Process Monitoring ; Business Process Management System ; Machine Learning ; Deep Learning ; Multi-Task Learning

 Digital Object List

 Bookmark

...see more