Scholarlink

<p>With the increasing adoption of cloud infrastructure for workflow process deployment, the workflow process remaining time prediction has received more attention due to its significance in improving response time and optimizing resource allocation. Event encoding techniques, by capturing the temporal dynamics and contextual dependencies of process events, improve the accuracy of workflow process remaining time predictions. However, there remains a lack of comprehensive empirical evaluation to analyze the impact of various event encoding techniques on prediction accuracy. To fill this gap, this paper conducts an extensive experimental evaluation of five state-of-the-art event encoding techniques, including One-Hot, Skip-Gram, CBOW (Continuous Bag-of-Word), FastText and GloVe, across nine prediction models based on LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), and QRNN (Quasi-Recurrent Neural Network). The evaluation utilizes eight real-world event logs to assess the accuracy of workflow remaining time predictions. The experimental results demonstrate that the GloVe encoding technique consistently yields superior prediction accuracy across the majority of prediction models and event logs.</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

Comparative evaluation of encoding techniques for workflow process remaining time prediction for cloud applications

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月3日

查看全文
  • Cong Liu,
  • Wenjuan Liu,
  • Na Guo,
  • Rongjia Song,
  • Yan Gu,
  • Long Cheng,
  • Qingtian Zeng

摘要

With the increasing adoption of cloud infrastructure for workflow process deployment, the workflow process remaining time prediction has received more attention due to its significance in improving response time and optimizing resource allocation. Event encoding techniques, by capturing the temporal dynamics and contextual dependencies of process events, improve the accuracy of workflow process remaining time predictions. However, there remains a lack of comprehensive empirical evaluation to analyze the impact of various event encoding techniques on prediction accuracy. To fill this gap, this paper conducts an extensive experimental evaluation of five state-of-the-art event encoding techniques, including One-Hot, Skip-Gram, CBOW (Continuous Bag-of-Word), FastText and GloVe, across nine prediction models based on LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), and QRNN (Quasi-Recurrent Neural Network). The evaluation utilizes eight real-world event logs to assess the accuracy of workflow remaining time predictions. The experimental results demonstrate that the GloVe encoding technique consistently yields superior prediction accuracy across the majority of prediction models and event logs.