基于可扩展活动关系的过程概念漂移检测

Process concept drift detection based on extensible activity relationship

为了发现过程模型漂移的时间点,提出一种基于活动关系频繁度的日志分割方法。事件日志可以用活动关系抽象表示。通过关系抽取将事件日志转化为活动关系矩阵,然后分析每个活动关系的变化趋势并检测出候选变更点将所有候选变更点通过密度聚类的方式进行合并,得到模型漂移的时间点。在人工生成日志上的实验结果表明,算法具有良好的准确率、较小的误差和较低的时间消耗。

To detect change points from event logs based on relation frequency, an approach was proposed with concept drift phenomenon in process mining. Event logs could be characterized by relationships between activities and transformed into a relation matrix. By analyzing the variation of activity relations, candidate change points were detected. The final result was calculated from candidate change points via DBSCAN. Experiments on synthetic logs showed that the proposed approach possessed high accuracy, low error and good time performance.