Scholarlink

<p>Time series prediction has been widely used in the medical field to predict patient recurrence or physiological fluctuations. However, the adequacy of the existing methods for contextual information interaction is still insufficient when dealing with a longer memory need in clinical data modelling. In order to enhance the utilization of memory interaction, this paper introduces a new contextual interaction refinement method MB-LSTM by incorporating a Hidden Layer Information Interaction Intensifier. The MB-LSTM method allows for simultaneous interaction of input and hidden layer states at each time step to enhance capability of capturing complex temporal relationships. Besides, more features of time series data are learned utilizing contrastive learning and a data augmentation scheme based on Kernel Density Estimation is designed to identify more accurate features from time series data. The method is evaluated on a real clinical dataset including 1053 records of patient with Gouty arthritis from the Guangdong Provincial Traditional Chinese Medicine Hospital by predicting the subsequent status of patients. The results show the proposed method achieves state-of-the-art performance by 0.5–7.2% using four different evaluation metrics compared with baseline methods.</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

MBLSTM is a contextual interaction refined method for time series prediction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年5月27日

查看全文
  • Weihan Qiu,
  • Fangjie Zhu,
  • Tianyong Hao,
  • Maojie Wang,
  • Runyue Huang

摘要

Time series prediction has been widely used in the medical field to predict patient recurrence or physiological fluctuations. However, the adequacy of the existing methods for contextual information interaction is still insufficient when dealing with a longer memory need in clinical data modelling. In order to enhance the utilization of memory interaction, this paper introduces a new contextual interaction refinement method MB-LSTM by incorporating a Hidden Layer Information Interaction Intensifier. The MB-LSTM method allows for simultaneous interaction of input and hidden layer states at each time step to enhance capability of capturing complex temporal relationships. Besides, more features of time series data are learned utilizing contrastive learning and a data augmentation scheme based on Kernel Density Estimation is designed to identify more accurate features from time series data. The method is evaluated on a real clinical dataset including 1053 records of patient with Gouty arthritis from the Guangdong Provincial Traditional Chinese Medicine Hospital by predicting the subsequent status of patients. The results show the proposed method achieves state-of-the-art performance by 0.5–7.2% using four different evaluation metrics compared with baseline methods.