Scholarlink

<p>Human emotions play an integral role in cognition and behavior, making emotion recognition a valuable tool for mental health monitoring. This work presents an interpretable EEG-based emotion recognition framework that integrates Explainable Artificial Intelligence (XAI) with advanced tree-based classifiers, including the Deep Forest Cascading Classifier (DFCC), XGBoost, and Decision Trees. The proposed approach employs SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) to identify key EEG features, such as mean_2_b, min_q_0_a, and min_q_15_b—that significantly influence classification decisions. This interpretability aligns the model’s decision-making process with established psychological theories, enhancing its trustworthiness for clinical applications. Experimental evaluation demonstrates DFCC’s superior performance, achieving 99.30% accuracy and an AUC of 0.9999 in One-vs-Rest settings, outperforming XGBoost (99.07%) and Decision Trees (96.79%). Paired t-tests (t(4) = 4.21, <i>p</i> = 0.012), McNemar’s test (χ²=6.78, <i>p</i> = 0.009), and a 95% confidence interval (98.5–99.8%) confirm the statistical reliability of these results.This research highlights the potential of combining high accuracy, interpretability, and statistical validation to develop clinically relevant and transparent systems, paving the way for reliable applications in mental health diagnostics and adaptive human–computer interaction (HCI) systems.</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

DeepXAI-ERS: An Explainable Deep Forest Approach for EEG-Based Emotion Recognition Systems

  • 期刊论文

  • 发表日期：

    2025年10月15日

查看全文
  • J. Vakala Rani,
  • Y. Swathi,
  • Prathiba V. Kalburgi,
  • Binaya Budhathoki,
  • Aishwarya Jakka

摘要

Human emotions play an integral role in cognition and behavior, making emotion recognition a valuable tool for mental health monitoring. This work presents an interpretable EEG-based emotion recognition framework that integrates Explainable Artificial Intelligence (XAI) with advanced tree-based classifiers, including the Deep Forest Cascading Classifier (DFCC), XGBoost, and Decision Trees. The proposed approach employs SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) to identify key EEG features, such as mean_2_b, min_q_0_a, and min_q_15_b—that significantly influence classification decisions. This interpretability aligns the model’s decision-making process with established psychological theories, enhancing its trustworthiness for clinical applications. Experimental evaluation demonstrates DFCC’s superior performance, achieving 99.30% accuracy and an AUC of 0.9999 in One-vs-Rest settings, outperforming XGBoost (99.07%) and Decision Trees (96.79%). Paired t-tests (t(4) = 4.21, p = 0.012), McNemar’s test (χ²=6.78, p = 0.009), and a 95% confidence interval (98.5–99.8%) confirm the statistical reliability of these results.This research highlights the potential of combining high accuracy, interpretability, and statistical validation to develop clinically relevant and transparent systems, paving the way for reliable applications in mental health diagnostics and adaptive human–computer interaction (HCI) systems.