Scholarlink

<p>Extracting knowledge from business process logs to prevent violations of business rules can protect companies from major losses. Most of the existing approaches toward this goal focus on compliance verification with respect to a target business model and are purely reactive: they detect violations <i>ex post</i>. The few existing approaches that try to prevent violations beforehand require substantial manual intervention, don’t consider fine-grained logs, ordinarily found in real-world business scenarios, and are based on memoryless techniques. To fill these gaps, we propose an integrated end-to-end framework to predict business model violations from a stream of low-level event logs. We use a Bidirectional Long-Short-Term Memory (BiLSTM) model, integrated with an attention mechanism to capture discriminating features and enable training on long sequences. This framework, whose setup requires minimal human intervention, can forecast not only the type but also the relative location of the upcoming violations in the event sequence. This information is useful in determining the type of countermeasures that need to be taken. We demonstrate the applicability of the framework using a real-life event log and achieve a prediction accuracy of 99.74%.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Deep Learning Framework for Predicting Business Process Violations

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月4日

查看全文
  • Ghalia Tello,
  • Gabriele Gianini,
  • Rabeb Mizouni,
  • Corrado Mio,
  • Ernesto Damiani,
  • Paolo Ceravolo

摘要

Extracting knowledge from business process logs to prevent violations of business rules can protect companies from major losses. Most of the existing approaches toward this goal focus on compliance verification with respect to a target business model and are purely reactive: they detect violations ex post. The few existing approaches that try to prevent violations beforehand require substantial manual intervention, don’t consider fine-grained logs, ordinarily found in real-world business scenarios, and are based on memoryless techniques. To fill these gaps, we propose an integrated end-to-end framework to predict business model violations from a stream of low-level event logs. We use a Bidirectional Long-Short-Term Memory (BiLSTM) model, integrated with an attention mechanism to capture discriminating features and enable training on long sequences. This framework, whose setup requires minimal human intervention, can forecast not only the type but also the relative location of the upcoming violations in the event sequence. This information is useful in determining the type of countermeasures that need to be taken. We demonstrate the applicability of the framework using a real-life event log and achieve a prediction accuracy of 99.74%.