Loading...

Data-Driven Prediction for Monitoring Business Process Pperformances Based on Classification Algorithms

Taheriyan, Zahra | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57421 (01)
  4. University: Sharif University of Technology
  5. Department: Industrial Engineering
  6. Advisor(s): Hassannayebi, Erfan
  7. Abstract:
  8. In recent years, several studies have been conducted in the field of data mining techniques in the field of process mining with the aim of improving the performance of organizations. Predictive process monitoring is a data-driven approach that helps business managers to improve the status and conditions of their organization. In this approach, the event log, which includes a set of completed examples of a process, is received as input, and machine learning methods are used to predict the outcome and results of the organization's processes before the process is completed. This prediction can include the prediction of the final result, the next event, the time remaining until the completion of the process. In this research, outcome-oriented predictive process monitoring has been used. The innovation of this research includes the use of the Catboost method in the well-known approach of predictive process monitoring and the use of this method along with inter-case features, along with the use of last-state, aggregation, and index-based encoding methods. Since improving the prediction of process results is one of the main goals of this research, various methods were used to improve the results. The proposed approach was implemented on the data of the traffic fine management process in Italy. The results showed that the use of Catboost algorithm in the famous approach of predictive process monitoring in previous research led to a 5% improvement in the F-score measurement and 3% AUC for last state encoding and a 9% improvement in F-score and 4% AUC for aggregation approach encoding in the first question of the investigation (payment or non-payment of the fine by the offender) is anticipated. By applying index-based encoding for this question, the result obtained a value of 0.77 for the F-score criterion and a value of 0.74 for the AUC index shows the better performance of this encoding function than the other two encoding functions. Also, the combination of this classification algorithm with inter-case features leads to a 6% improvement in the F-score criterion and a 4% improvement in the AUC for the last-state encoding, an 8% improvement in the F-score index and a 5% AUC improvement, and The F-score criterion was improved by 9% and the AUC index by 3% in predicting the second question of this research (determining the time range of the province's response to the Italian local police). This features include the number of samples that have started but not finished, the number of samples that have not been finished and their last event was sending a complaint to the province and the rate of creation and completion of the number of new files since the start of the current file. This issue shows the improvement of the accuracy of the outcome-oriented predictive process monitoring compared to other well-known classical machine learning methods, including XGBoost, one of the best of these methods
  9. Keywords:
  10. Process Mining ; Predictive Process Monitoring ; Data Mining ; Data Driven Method ; Machine Learning

 Digital Object List

 Bookmark