Scholarlink

<p>Electroencephalography (EEG) based Emotion recognition is a fundamental problem in the field of affective computing, being non-stationary is a major obstacle in brain signal processing. Although Echo State Networks (ESNs) are quite successful in managing time sequences of data, their efficacy is bound by the fact of randomly initialized reservoir connections which are not quite good for the task at hand. This paper puts out a completely different framework to fix this problem with an optimum algorithm of the reservoir topology, for the ESN to be used in the recognition of emotions from EEG. Our method evaluates and refines the internal connections of the reservoir to maximize its predictive performance on emotion classification tasks. Furthermore, we integrate explainable artificial intelligence (XAI) tools—Local Interpretable Model-agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP)—for the purposes of model decision recording and identifying the most dominant EEG features. Extensive experimental results on two benchmark datasets, DEAP and SEED, demonstrate the superiority of our approach. Our method achieves 93.79% classification accuracy on SEED and 92.4% valence classification on DEAP, both of which are the best results among all existing studies. The proposed optimized ESN framework proves to be a powerful, accurate, and interpretable tool for advancing emotion recognition from EEG signals.</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

Emotion recognition with EEG signals using an insightful adapted echo state network framework

  • 期刊论文

  • 发表日期：

    2025年11月21日

查看全文
  • Samar Bouazizi,
  • Hela Ltifi

摘要

Electroencephalography (EEG) based Emotion recognition is a fundamental problem in the field of affective computing, being non-stationary is a major obstacle in brain signal processing. Although Echo State Networks (ESNs) are quite successful in managing time sequences of data, their efficacy is bound by the fact of randomly initialized reservoir connections which are not quite good for the task at hand. This paper puts out a completely different framework to fix this problem with an optimum algorithm of the reservoir topology, for the ESN to be used in the recognition of emotions from EEG. Our method evaluates and refines the internal connections of the reservoir to maximize its predictive performance on emotion classification tasks. Furthermore, we integrate explainable artificial intelligence (XAI) tools—Local Interpretable Model-agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP)—for the purposes of model decision recording and identifying the most dominant EEG features. Extensive experimental results on two benchmark datasets, DEAP and SEED, demonstrate the superiority of our approach. Our method achieves 93.79% classification accuracy on SEED and 92.4% valence classification on DEAP, both of which are the best results among all existing studies. The proposed optimized ESN framework proves to be a powerful, accurate, and interpretable tool for advancing emotion recognition from EEG signals.