Scholarlink

<p>This paper presents a novel, explainable feature engineering framework for classifying EEG and ECG signals with high accuracy. The proposed method employs the Order Transition Pattern (OTPat) feature extractor. The presented OTPat feature extractor captures both channel/column-based patterns (spatial features) using all channels for each point and signal/row-based patterns (temporal features) by extracting features from individual channels using overlapping blocks. The extracted features are then refined using cumulative weighted iterative neighborhood component analysis (CWINCA) for feature selection and classified with a t‑algorithm k‑nearest neighbors (tkNN) classifier. Finally, two symbolic languages, Directed Lobish (DLob) and Cardioish, generate interpretable results in the form of cortical and cardiac connectome diagrams. The OTPat-based XFE model achieves over 95% accuracy on several EEG and ECG datasets and reaches 86.07% accuracy on an 8‑class EEG artifact dataset. These results demonstrate high performance and clear interpretability, highlighting the model’s potential for robust biomedical signal classification.</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

Novel accurate classification system developed using order transition pattern feature engineering technique with physiological signals

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年5月1日

查看全文
  • Mehmet Ali Gelen,
  • Prabal Datta Barua,
  • Irem Tasci,
  • Gulay Tasci,
  • Emrah Aydemir,
  • Sengul Dogan,
  • Turker Tuncer,
  • U. R. Acharya

摘要

This paper presents a novel, explainable feature engineering framework for classifying EEG and ECG signals with high accuracy. The proposed method employs the Order Transition Pattern (OTPat) feature extractor. The presented OTPat feature extractor captures both channel/column-based patterns (spatial features) using all channels for each point and signal/row-based patterns (temporal features) by extracting features from individual channels using overlapping blocks. The extracted features are then refined using cumulative weighted iterative neighborhood component analysis (CWINCA) for feature selection and classified with a t‑algorithm k‑nearest neighbors (tkNN) classifier. Finally, two symbolic languages, Directed Lobish (DLob) and Cardioish, generate interpretable results in the form of cortical and cardiac connectome diagrams. The OTPat-based XFE model achieves over 95% accuracy on several EEG and ECG datasets and reaches 86.07% accuracy on an 8‑class EEG artifact dataset. These results demonstrate high performance and clear interpretability, highlighting the model’s potential for robust biomedical signal classification.