Scholarlink

<p>The inherent noise and high dimensionality of EEG signal make it more difficult to accomplish fast and accurate classification, which is particularly challenging in rapid serial visual presentations (RSVP) target recognition tasks based on EEG. Conventional feature selection approaches have difficulty in reducing dimensions and keeping a good classification performance at the same time. To resolve these issues, we present a bio-inspired hybrid optimization framework where Differential Evolution (DE) is integrated with Grey Wolf Optimization (GWO) to improve the efficiency of feature selections. Our method utilizes the focal loss for combating data imbalance in EEG datasets, the Fisher score for enhancing the discrimination capability of the classes, and the average pairwise Pearson correlation for the reduction of redundant information between features in RSVP datasets. In such a sense, GWO functions as an exploration tool of the feature space, while DE further refines good candidates through a local exploitation. CFSFs are tested with a variety of classifiers ranging from traditional classifiers to deep learning, such as support vector machine (SVM), light gradient boosting machine (LightGBM), convolutional neural network (CNN), long short-term memory (LSTM), etc. Compared with those traditional optimization and classification pipelines, our hybrid GWO–DE method converges faster displacement, more efficient dimension reduction, higher detection accuracy, and classification accuracy, and F1 score result showed us 98.79% and 97.7%, respectively. This efficient method is applicable to real-time EEG applications, including cognitive monitoring, neuro-feedback, and biometric identification.</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

An Enhanced Bio-inspired GWO–DE Technique for Efficient Feature Selection in the EEG-RSVP Paradigm

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月6日

查看全文
  • S. Abinayaa,
  • S. S. Sridhar

摘要

The inherent noise and high dimensionality of EEG signal make it more difficult to accomplish fast and accurate classification, which is particularly challenging in rapid serial visual presentations (RSVP) target recognition tasks based on EEG. Conventional feature selection approaches have difficulty in reducing dimensions and keeping a good classification performance at the same time. To resolve these issues, we present a bio-inspired hybrid optimization framework where Differential Evolution (DE) is integrated with Grey Wolf Optimization (GWO) to improve the efficiency of feature selections. Our method utilizes the focal loss for combating data imbalance in EEG datasets, the Fisher score for enhancing the discrimination capability of the classes, and the average pairwise Pearson correlation for the reduction of redundant information between features in RSVP datasets. In such a sense, GWO functions as an exploration tool of the feature space, while DE further refines good candidates through a local exploitation. CFSFs are tested with a variety of classifiers ranging from traditional classifiers to deep learning, such as support vector machine (SVM), light gradient boosting machine (LightGBM), convolutional neural network (CNN), long short-term memory (LSTM), etc. Compared with those traditional optimization and classification pipelines, our hybrid GWO–DE method converges faster displacement, more efficient dimension reduction, higher detection accuracy, and classification accuracy, and F1 score result showed us 98.79% and 97.7%, respectively. This efficient method is applicable to real-time EEG applications, including cognitive monitoring, neuro-feedback, and biometric identification.