An Anti-Noise Process Mining Algorithm Based on Minimum Spanning Tree Clustering

• Published in: IEEE Access Vol. 6; pp. 48756 - 48764
• Main Authors: Li, Weimin, Zhu, Heng, Liu, Wei, Chen, Dehua, Jiang, Jiulei, Jin, Qun
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.01.2018; Institute of Electrical and Electronics Engineers (IEEE); The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Analytical models; BPM; business process; Business process management; Clustering; Clustering algorithms; Data mining; Electrical engineering. Electronics. Nuclear engineering; event log; Graph theory; Management systems; Noise; Noise reduction; Petri nets; Process mining; Task analysis; TK1-9971; Transforms
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2018.2865540

Many human-centric systems have begun to use business process management technology in production. With the operation of business process management systems, more and more business process logs and human-centric data have been accumulated. However, the effective utilization and analysis of these event logs are challenges that people need to solve urgently. Process mining technology is a branch of business process management technology. It can extract process knowledge from event logs and build process models, which helps to detect and improve business processes. The current process mining algorithms are inadequate in dealing with log noise. The family of alpha-algorithms ignores the impact of noise, which is unrealistic in real-life logs. Most of the process mining algorithms that can handle noise also lack reasonable denoising thresholds. In this paper, a new assumption on noise is given. Furthermore, an anti-noise process mining algorithm that can deal with noise is proposed. The decision rules of the selective, parallel, and non-free choice structures are also given. The proposed algorithm framework discovers the process model and transforms it into a Petri network representation. We calculate the distance between traces to build the minimum spanning tree on which clusters are generated. The traces of the non-largest clusters are treated as noise, and the largest cluster is mined. Finally, the algorithm can discover the regular routing structure and solve the problem of noise. The experimental results show the correctness of the algorithm when compared with the <inline-formula> <tex-math notation="LaTeX">\alpha </tex-math></inline-formula>++ algorithm.