Appendix A. Overview of Methods and Results for Emotion Recognition Studies

Ref
Paper Title
Target Variables for DREAMER
Key Results
Signal Type
Dataset Used
Year
DL/ML Techniques Used

]

Embedded EEG Feature Selection for Multi-Dimension Emotion Recognition

Multi-dimensional labels: valence, arousal, dominance (8-class)

86.43% accuracy

EEG

DREAMER, DEAP, HDED

2024

Feature Selection Methods

Fractal Spiking Neural Network Scheme for EEG-Based Emotion Recognition

Valence, Arousal, Dominance (binary: high vs. low)

SD: Valence 71.01%, Arousal 78.5%, Dominance 80.92%; SI: Arousal 70.53%, Dominance 75.12%

EEG

DREAMER, DEAP, SEED-IV, MPED

2024

Spiking Neural Network

A Hybrid End-to-End Spatiotemporal Attention Neural Network With Graph-Smooth Signals for EEG Emotion Recognition

Valence, Arousal (binary: high vs. low)

Transfer Learning improved valence to 83% and arousal to 87.2% with 90% DREAMER data

EEG

DEAP, DREAMER, EEWD

2024

Spatiotemporal Attention Network

LUR: An Online Learning Model for EEG Emotion Recognition

Valence (positive/negative), Arousal (high/low)

Arousal: 71%, Valence: 66%; Valence accuracy per subject up to 90%

EEG

DEAP, DREAMER

2023

SVM, Naive Bayes, Online Learning

EEG-Based Emotion Feature Extraction Using Power Spectral Density

Emotion States: Happy (High Valence), Sad (Low Valence)

PSD calculations across theta, alpha, beta bands showed strong correlations for valence (Happy vs. Sad); specific EEG channels showed higher PSD for "Happy" state

EEG

DREAMER

2023

-

Robust Emotion Recognition in EEG Signals Based on a Combination of Multiple Domain Adaptation Techniques

Valence: Positivity/Negativity, Arousal: Excitement/Calmness

MDA-NF: Valence 74.5%, Arousal 71.5%; Cross-dataset: DREAMER→DEAP, valence accuracy 64.45%

EEG

DEAP, DREAMER

2023

MDA-NF technique

EEG-Based Emotion Analysis Using Person-Event Network

Valence (Positivity/Negativity), Arousal (Intensity)

Valence accuracy: 83.27%, Arousal accuracy: 88.64%; Better performance with combined person and event-related features

EEG

DEAP, DREAMER

2023

CNN

CycleMVAE: Benchmarking End-to-End Cycle-Consistent Multi-Task Variational Autoencoder for EEG-Based Emotion Recognition

Arousal, Valence, Dominance (High/Low)

F1-Score: Arousal 0.69, Valence 0.68, Dominance 0.71; Accuracy: 78.21% - 80.78%

EEG

DREAMER

2023

VAE

A Hierarchical Three-Dimensional MLP-Based Model for EEG Emotion Recognition

Valence (Positivity/Negativity), Arousal (Excitement/Calmness)

Accuracy: Valence 62.51%, Arousal 64.49%; Hierarchical structure enhances noise robustness

EEG

DEAP, DREAMER, SEED-IV

2023

MLP

EVNCERS: An Integrated Eigenvector Centrality-Variational Nonlinear Chirp Mode Decomposition-Based EEG Rhythm Separation for Automatic Emotion Recognition

Valence, Arousal, Dominance (High/Low); threshold 3 on 1-5 scale

Valence: 62.51%, Arousal: 64.49%; Highlights include using Eigenvector Centrality to enhance feature stability

EEG

DREAMER, SEED

2023

ML classifiers

Emotion Recognition based on fusion of multimodal physiological signals using LSTM and GRU

Valence, Arousal, Dominance (High/Low); threshold 3 on 1-5 scale

Arousal: 96.86% accuracy (F1: 97.98%), Valence: 96.87% accuracy (F1: 97.82%), Dominance: 96.81% accuracy (F1: 97.71%); Delta rhythm shows highest prediction accuracy

EEG, ECG, respiration, temperature

SEED, DREAMER, WESAD

2023

LSTM, GRU, 1DCNN

EEG-Based Emotion Recognition Using Trainable Adjacency Relation Driven Graph Convolutional Network

Valence, Arousal, Dominance (Sad, Neutral, Happy classes)

SVM/Random Forest accuracy: 50-60%; LSTM: 65%, GRU: 62% –

EEG

DREAMER, DEAP

2023

GCN

Three Dimensional Emotion State Classification based on EEG via Empirical Mode Decomposition

Valence, Arousal, Dominance (High/Low); threshold 3 on 1-5 scale

IMF4 provides best accuracy; Valence: 78.32%, Arousal: 76.22%, Dominance: 77.22%;

EEG

AMIGOS, DREAMER

2023

FFNN

EEG evoked automated emotion recognition using deep convolutional neural network

Valence, Arousal, Dominance (High/Low); 1-3 low, 4-5 high

DCNN Accuracy: Valence 84%, Arousal 94.4%, Dominance 97.6%; Outperforms RF and XGBoost on DREAMER

EEG

DREAMER

2023

DCNN

FedEmo: A Privacy-Preserving Framework for Emotion Recognition using EEG Physiological Data

Valence, Arousal, Dominance (High/Low; threshold >3 high)

Valence: 63.3%, Arousal: 56.7%, Dominance: 52.2%;

EEG

DREAMER

2023

ANN

Bi-CapsNet: A Binary Capsule Network for EEG-Based Emotion Recognition

Valence, Arousal, Dominance (Low ≤3, High >3)

Bi-CapsNet showed <1% accuracy drop vs. FP-CapsNet; achieved high accuracy, lower computation costs, and 5x speed on mobile vs. FP-CapsNet

EEG

DEAP, DREAMER

2023

Binary capsule network

A Novel Tensorial Scheme for EEG-Based Person Identification

Person Identification

Study on person identification accuracy via EEG features on DREAMER and DEAP datasets

EEG

DEAP, SEED, DREAMER

2023

-

SSTD: A Novel Spatio-Temporal Demographic Network for EEG-Based Emotion Recognition

Valence, Arousal (Binary; median 3)

Valence: 76.81%, Arousal: 81.64%;

EEG

DEAP, DREAMER

2023

GRU, SPDNet

Residual GCB-Net: Residual Graph Convolutional Broad Network on Emotion Recognition

Arousal, Valence, Dominance (Binary; Low/High)

Arousal: 91.55%, Dominance: 89.37%, Valence: 87.43%;

EEG

SEED, Dreamer

2023

Residual learning blocks, Graph convolutional broad network

EEG-Based Emotion Recognition via Neural Architecture Search

Arousal, Valence, Dominance (Binary; High/Low with threshold of 3)

Arousal: 96.62%, Valence: 96.29%, Dominance: 96.61%

EEG

DEAP, DREAMER

2023

Neural architecture search, Reinforcement learning

AT2GRU: A Human Emotion Recognition Model With Mitigated Device Heterogeneity

Valence, Arousal, Dominance (Multi-class; 1-5 scale)

Valence: 84.15%, Arousal: 84.69%, Dominance: 85.81%;

EEG, ECG

DREAMER, DEAP

2023

Wavelet filters, GRU

Variational Instance-Adaptive Graph for EEG Emotion Recognition

Valence, Arousal (Binary; Low/High)

Valence: 92.82%, Arousal: 93.09% using subject-dependent evaluation

EEG

SEED, MPED, DREAMER

2023

Multi-level and multi-graph convolution operation

SparseDGCNN: Recognizing Emotion From Multichannel EEG Signals

Valence, Arousal (Binary; Positive/Negative, High/Low)

Subject-Dependent: Avg. 12.66% accuracy and 12.45% F1 improvement; Subject-Independent: Avg. 4.69% accuracy and 8.51% F1 improvement

EEG

SEED, DEAP, DREAMER, CMEED

2023

Forward-backward splitting method

EEG-Based Emotion Recognition via Channel-Wise Attention and Self Attention

Valence, Arousal, Dominance (Binary; High/Low)

Valence: 97.93%, Arousal: 97.98%, Dominance: 98.23% with ACRNN model

EEG

DEAP, DREAMER

2023

Channel-wise attention, Self-attention

Time-Varying Graph Signal Processing Based Cross-Subject Emotion Classification from Multi-Electrode EEG Signals

Valence, Arousal (Binary; Positive/Negative, High/Low with threshold 3)

Valence: 97.83%, Arousal: 93.56%; 10-fold cross-validation used

EEG

DREAMER

2022

KNN classifier

EEG-based Emotion Identification using General Factor Analysis

Valence, Arousal, Dominance (Binary; Low <3, High ≥3)

Valence: 96.8%, Arousal: 94.2%, Dominance: 92.6% with SVM

EEG

DREAMER

2022

SVM, KNN

Exploiting Multiple EEG Data Domains with Adversarial Learning

Discrete Emotion States (Negative, Neutral, Positive)

Multi-label classification: 40.48% accuracy; Binary classification (Pos/Neg): 58.17% accuracy

EEG

SEED, SEED-IV, DEAP, DREAMER

2022

-

Joint Temporal Convolutional Networks and Adversarial Discriminative Domain Adaptation for EEG-Based Cross-Subject Emotion Recognition

Arousal, Valence (Binary; High/Low via k-means)

Valence: 66.56%, Arousal: 63.69%;

EEG

DEAP, DREAMER

2022

Temporal Convolutional Networks (TCN), Adversarial Discriminative Domain Adaptation (ADDA)

Ensemble Machine Learning-Based Affective Computing for Emotion Recognition Using Dual-Decomposed EEG Signals

Arousal, Valence (Multiclass)

Arousal: 95.81%, Valence: 95.53%;

EEG

DREAMER, SEED, INTERFACES, MUSEC

2022

Ensemble Machine Learning Algorithms

Graph-Embedded Convolutional Neural Network for Image-Based EEG Emotion Recognition

Valence and Arousal (Binary; High/Low)

Valence: 95.73%, Arousal: 92.79%

EEG

SEED, SDEA, DREAMER, MPED

2022

GECNN with attention mechanism and dynamical graph filtering

Comparing Recognition Performance and Robustness of Multimodal Deep Learning Models

Arousal, Valence, Dominance (Binary; High/Low)

Arousal: 89%, Valence: 90.6%, Dominance: 90.7%

EEG, Eye Movement

SEED, SEED-IV, DEAP, SEED-V, DREAMER

2022

DCCA with weighted sum and attention-based fusion

Advancing Remote Healthcare Using Humanoid and Affective Systems

Valence, Arousal (Binary; High >3, Low ≤3)

Capsule Network Accuracy: 90.4%

ECG, GSR

DREAMER, AMIGOS

2022

Deep Learning Algorithms

GCB-Net: Graph Convolutional Broad Network and Its Application in Emotion Recognition

Valence, Arousal, Dominance (Binary; High >2, Low ≤2)

Valence: 86.99%, Arousal: 89.32%, Dominance: 89.20%

EEG

SEED, DREAMER

2022

Graph Convolutional Broad Network (GCB-Net)

Utilizing Deep Learning Towards Multi-Modal Bio-Sensing and Vision-Based Affective Computing

Valence, Arousal (Binary; High/Low)

Valence: 79.95%, Arousal: 79.95%, Emotion Classification: 55.56%

EEG, ECG, GSR, PPG

DEAP, MAHNOB-HCI, AMIGOS, DREAMER

2022

Deep Learning Algorithms

Human Emotion Recognition using EEG Signal in Music Listening

Discrete emotion states (9 classes)

k-NN: 94.49%, RF: 99.94%, XGBoost: 99.39%, CNN: 99.91%

EEG

DREAMER

2021

k-NN, Random Forest, XGBoost, CNN

Evaluation of TEAP and AuBT as ECG's Feature Extraction Toolbox

Valence, Arousal (Binary; High/Low)

TEAP (Arousal): 65.4%, AuBT (Valence): 65.8%

ECG

DREAMER, AuBT dataset

2021

Support Vector Machine (SVM)

Feature Reconstruction Based Channel Selection for Emotion Recognition Using EEG

Valence (Binary; Positive/Negative)

Baseline AUC with 14 channels: 0.83, AUC with 8 channels: 0.80

EEG

AMIGOS, DREAMER, DEAP

2021

Machine Learning Models

Online Cross-subject Emotion Recognition from ECG via Unsupervised Domain Adaptation

Valence, Arousal (Binary; Positive/Negative, High/Low)

Valence Accuracy: 72% (F1: 0.69), Arousal Accuracy: 71% (F1: 0.65)

ECG

Dreamer, Amigos

2021

Unsupervised Domain Adaptation

A Computerized Approach for Automatic Human Emotion Recognition Using Sliding Mode Singular Spectrum Analysis

Arousal, Dominance, Valence

(Binary; High/Low)

KNN: Arousal 92.06%, Dominance 92.38%, Valence 92.3%; SVM: Arousal 92.01%, Dominance 92.29%, Valence 91.53%

EEG, ECG

DREAMER, AMIGOS

2021

Machine Learning Classifiers

A Novel Spatio-Temporal Field for Emotion Recognition Based on EEG Signals

Valence, Arousal

(Binary: High vs. Low)

Valence: 81.2%, Arousal: 82.4%

EEG

DEAP, DREAMER

2021

-

eRAD-Fe: Emotion Recognition-Assisted Deep Learning Framework

Valence, Arousal (Four Classes: HVHA, LVHA, HVLA, LVLA)

LSTM Accuracy: 81.51%; ET-MCSP feature extraction, 10-sec window with 30% overlap

EEG

SEED, DEAP, DREAMER

2021

Long Short-Term Memory (LSTM)

Attention-based Spatio-Temporal Graphic LSTM for EEG Emotion Recognition

Valence, Arousal

(Binary Classification)

EEG

DEAP, DREAMER

2021

Long Short-term Memory (LSTM), Graph Convolution

EEG Emotion Recognition Based on 3-D Feature Representation and Dilated Fully Convolutional Networks

Valence, Arousal, Dominance

(Binary: Low vs. High)

Subject-dependent: Valence 76.88%, Arousal 75.92%, Dominance 77.43%; LOSO cross-validation

EEG

DEAP, DREAMER

2021

Convolutional Neural Networks

FLDNet: Frame-Level Distilling Neural Network for EEG Emotion Recognition

Valence, Arousal, Dominance (Binary)

Valence: 89.91%, Arousal: 87.67%, Dominance: 90.28%

EEG

DEAP, DREAMER

2021

Neural Network

A Novel Multivariate-Multiscale Approach for Computing EEG Spectral and Temporal Complexity for Human Emotion Recognition

Valence, Arousal, Dominance

(Binary: High vs. Low)

Valence: 86.2%, Arousal: 84.5%, Dominance: 83.9%

EEG

SJTU emotion EEG dataset (SEED), DREAMER emotion EEG public database

2021

Sparse autoencoder, Random forest

Emotion Recognition From Multi-Channel EEG via Deep Forest

Valence, Arousal, Dominance

(Binary: High >3, Low ≤3)

Valence: 89.03%, Arousal: 90.41%, Dominance: 89.89%

Multi-channel EEG

DEAP, DREAMER

2021

Deep forest

BioCNN: A Hardware Inference Engine for EEG-Based Emotion Detection

Valence, Arousal, Dominance

(Binary)

Valence: 89.03%, Arousal: 90.41%, Dominance: 89.89%

EEG

DEAP, DREAMER, In-house dataset

2020

Convolutional Neural Network

EEG feature learning with Intrinsic Plasticity based Deep Echo State Network

Valence, Arousal, Dominance

(Binary: High ≥3, Low <3)

Valence: 82.11%, Arousal: 83.58%, Dominance: 84.98%

EEG

DEAP, DREAMER

2020

Deep echo state network, Intrinsic plasticity, Ridge regression, Online delta rule

EEG Emotion Recognition Using Dynamical Graph Convolutional Neural Networks

Valence, Arousal, Dominance

(Binary: High ≥3, Low <3)

Valence: 86.23% (SD 12.29%), Arousal: 84.54% (SD 10.18%), Dominance: 85.02% (SD 10.25%)

Multi-channel EEG

SEED, DREAMER

2020

DGCNN

Last updated