Abstract: Various anomalies may occur during the execution of business processes,which may pose risks to business organizations and lead to significant losses.To detect anomalous traces in event logs and locate anomalous activities within these traces,an ensemble learning-based framework combining a heuristic miner algorithm and autoencoders was proposed.A heuristic miner algorithm was employed to mine the process model and extract the main flows.Based on these main flows,the event logs were subjected to overlapping sampling,and autoencoders were trained for each sub-event log.If a particular trace failed to match any of the main flows or was detected as an anomaly by all autoencoders,it would be classified as an anomalous trace.Furthermore,by analyzing the anomalous traces along with their matched main flows,the specific activities causing the anomalies could be identified,leading to further measures for remediation or optimization.Experimental results demonstrated the effectiveness of this framework in efficiently detecting anomalies within business processes and accurately localizing anomalous activities within the traces.

Key words: business process, anomaly detection, ensemble learning, process mining, event log, autoencoder

摘要： 在业务流程执行中,可能会出现各种异常情况,从而给企业组织带来风险,导致巨大的损失。为了检测事件日志中的异常轨迹,并定位轨迹中的异常活动,提出一种结合启发式挖掘算法和自编码器模型的集成学习框架。首先,使用启发式挖掘算法来挖掘流程模型并提取主干。基于主干对事件日志进行重叠采样,并针对每个子事件日志训练自编码器模型。若某个轨迹无法匹配任何一条主干,或者被所有自编码器模型检测为异常,则该轨迹将被检测为异常。此外,通过对异常轨迹与其匹配的主干进行分析,可以确定引起异常的具体活动,并进一步采取相应的措施进行修复或优化。实验证明,该方法能够高效地检测业务流程中的异常,并能有效地定位轨迹中的异常活动。

关键词: 业务流程, 异常检测, 集成学习, 流程挖掘, 事件日志, 自编码器