EEG-Based Emotion Recognition for Machine Intelligence: A Survey with an Open-Source Python Toolbox

Clicks: 1
ID: 323027
2026
Free PDF
Article Quality & Performance Metrics
Overall Quality
Not rated
Combines reader engagement with the AI quality analysis. This article has not been analysed, so there is no overall score — reader engagement is measured and shown alongside.
AI Quality Assessment
Not analyzed
Readership in this journal

Ranked #246 of 267 articles by views in national science review

Most read Least read

Bar heights use a square-root scale. Only the 120 most-read articles are drawn; the journal has 267 in total.

Mint this article as an NFT
Not yet minted

Create a permanent, verifiable on-chain record of this article on the Scimatic Network. The NFT is held in your Journament account, and you can withdraw it to your own wallet at any time.

5 SUSD one-off · no wallet required
Sign in to mint
Abstract
Abstract This paper presents a survey of Electroencephalogram (EEG)-based emotion recognition, a task that utilizes the brain's electrical activity to decode genuine emotions. As a crucial research area in machine emotional intelligence, EEG-based emotion recognition (EEG-ER) has received extensive attention. While recent resources like TorchEEG and LibEER provide valuable coding libraries, they largely treat EEG-ER as a purely data-driven task, leaving a critical gap in bridging neurophysiological theory with computational pipeline design. To address this, we introduce an open-source Python toolbox, EEGEmoLib, accompanied by a methodological survey that explicitly integrates psychological and neuroscientific priors into algorithm design. Unlike existing reviews, we provide a structurally quantified framework incorporating 8 diverse public datasets and implementing 18 categories of handcrafted features with adaptive feature-weighting modules, 11 feature selection methods, 5 representative traditional machine learning models and 15 representative deep learning models spanning convolutional, sequential, graph-based, and Transformer architectures. Using this unified codebase, we conduct rigorous benchmarking under standardized subject-dependent and subject-independent protocols. Our comparative analysis suggests two concrete methodological insights: (1) explicitly modeling neurophysiological priors, such as hemispheric asymmetry and specific frequency band importance, can improve recognition performance in certain settings, and the effect is model-dependent rather than universal; and (2) while large-capacity Transformer models usually perform strongly in subject-dependent scenarios, their performance may degrade substantially in subject-independent settings, where shallower networks or explicit domain-adaptation models can exhibit better robustness. Detailed experimental results, methodological pitfalls, and outstanding issues are discussed to guide future research. The Python toolbox EEGEmoLib is publicly available at https://eegemolib.github.io.
Reference Key
openalex_W7171807206 Use this key to autocite in the manuscript while using SciMatic Manuscript Manager or Thesis Manager
Authors Yong‐Jin Liu, Kairui Wen, Yezhi Shu, Fang Liu, Ming Li, Pei Yang, Wenqi Ji, Chao Zhou, Huan Liu, Qinghua Zheng
Journal national science review
Year 2026
DOI
10.1093/nsr/nwag460
URL
https://doi.org/10.1093/nsr/nwag460
Keywords Keywords not found

Citations

No citations found. To add a citation, contact the admin at info@scimatic.org

Comments

No comments yet. Be the first to comment on this article.