Scholarlink

<p>Major depressive disorder (MDD) is a prevalent mental health disorder that has rapidly affected many people globally over the past decade. This paper introduces a lightweight Machine Learning framework by optimizing the number of significant EEG channels for automatic MDD detection. The proposed pipeline combines Binary Particle Swarm Optimization (BPSO) and Support Vector Machine (SVM). It suggests the four best channels: C4, Cz, F4, and T6, which are crucial for MDD detection. Heterogeneous biomarkers-including band powers, relative alpha, mean amplitude, Higuchi Fractal Dimension (HFD), Lempel–Ziv Complexity (LZC), Hurst Exponent, and Hjorth Complexity parameters-are extracted from preprocessed EEG-based MDD data. A baseline SVM model with all 19 channels achieves 91.38% average accuracy. The proposed BPSO-SVM framework, which uses only four EEG channels, improves accuracy by +5.17% (96.55%) under subject-independent LOSO cross-validation. In addition to these classification improvements and about 79% reduction in the number of channels, BPSO-SVM provides a lightweight and computationally efficient framework for real-time MDD detection. Statistical tests validate the consistent performance improvements of BPSO-SVM over the baseline SVM model. The BPSO-SVM model also outperforms recently introduced state-of-the-art approaches in terms of accuracy and the number of EEG channels required for MDD detection.</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

Binary Particle Swarm Optimization Based EEG Channel Selection for Major Depressive Disorder Detection

  • 期刊论文

  • 发表日期：

    2026年6月18日

查看全文
  • Divya Pachauri,
  • Urvashi Prakash Shukla,
  • Rahul Kumar Vijay

摘要

Major depressive disorder (MDD) is a prevalent mental health disorder that has rapidly affected many people globally over the past decade. This paper introduces a lightweight Machine Learning framework by optimizing the number of significant EEG channels for automatic MDD detection. The proposed pipeline combines Binary Particle Swarm Optimization (BPSO) and Support Vector Machine (SVM). It suggests the four best channels: C4, Cz, F4, and T6, which are crucial for MDD detection. Heterogeneous biomarkers-including band powers, relative alpha, mean amplitude, Higuchi Fractal Dimension (HFD), Lempel–Ziv Complexity (LZC), Hurst Exponent, and Hjorth Complexity parameters-are extracted from preprocessed EEG-based MDD data. A baseline SVM model with all 19 channels achieves 91.38% average accuracy. The proposed BPSO-SVM framework, which uses only four EEG channels, improves accuracy by +5.17% (96.55%) under subject-independent LOSO cross-validation. In addition to these classification improvements and about 79% reduction in the number of channels, BPSO-SVM provides a lightweight and computationally efficient framework for real-time MDD detection. Statistical tests validate the consistent performance improvements of BPSO-SVM over the baseline SVM model. The BPSO-SVM model also outperforms recently introduced state-of-the-art approaches in terms of accuracy and the number of EEG channels required for MDD detection.