Biography

Anomaly detection in business processes is crucial to prevent erroneous organizational decision-making. Existing anomaly detection methods often assume access to a clean event log without anomalies, a scenario that is impractical in the real world. To solve this problem, MuGAD is proposed, a novel framework that detects anomalies in scenarios where only a limited number of labeled anomalies are available. By converting traces into a newly designed graph structure, the framework captures the control flow, order of events, and attribute information, facilitating the effective use of graph neural networks. Furthermore, MuGAD uses a two-stage training procedure designed to (1) reduce the risks posed by unlabeled traces and (2) enable the detection of anomalous traces and events. Comparative experiments are conducted on five real-life event logs to validate our proposed framework. The experimental results indicate that MuGAD outperforms the existing methods and provides better detection accuracy along with better interpretability in terms of F1-score.

In a container terminal, the length of time that containers remain in the yard, known as Container Dwell Time (CDT), is considered one of the significant operational indicators due to its direct correlation with terminal productivity and efficiency. However, due to complex processing procedure and the involvement of various logistics stakeholders, CDT is subject to high uncertainty, making it more difficult for the terminal to manage. To address this issue, this paper presents a comprehensive framework to identify the Key Factors (KFs) influencing prolongation of CDT for import containers. In order to elucidate abnormal cases from dataset which contains yard loading information, the Process Mining (PM) method is utilized. Subsequently, XAI has been utilized to identify the KFs of import CDT. To reflect reality as closely as possible, we collected event data from a container terminal in Busan, Korea. Based on experiments, the KFs thus identified were: 1) Temperature, 2) Weight of container, 3) Voyage number of container 4) Block, 5) Shipping company, and 6) Month of discharging. To conclude, we formulated domain knowledge-based interpretations of the six most influential KFs.

Recent advancements in predictive process monitoring (PPM) have led to an improved performance by incorporating deep learning methodologies. However, some of these approaches employ a massive parameter architecture that requires a large dataset for training, posing challenges in many business process management scenarios. This study presents a transformer-based model with transfer learning for estimating remaining time prediction. We aim to improve the efficiency of business process management by accurately predicting the remaining time of a process. Additionally, we assessed the feasibility and transferability of the transfer learning method to enhance the performance of predictive process monitoring. We conducted experiments on four publicly available event logs. Our proposed architecture outperforms prior work, with a significant 53% improvement over the best baseline in the Helpdesk dataset. Additionally, the use of transfer learning leads to both positive and negative performance outcomes, depending on characteristics of the source and target process model.