Enhanced cross-dataset electroencephalogram-based emotion recognition using unsupervised domain adaptation

This repository contains the code accompanying our paper Enhanced Cross-Dataset Electroencephalogram-based Emotion Recognition using Unsupervised Domain Adaptation published in Computers in Biology and Medicine. Our work presents a domain-adaptive deep network for EEG-based emotion classification, aiming to improve cross-domain model performance by addressing feature distribution discrepancies. We introduce a sample selection technique to reduce negative transfer and propose a cost-effective test-time augmentation method to enhance test performance.

Setup Environment

The experiments were conducted using the following environment and packages:

• PyTorch == 2.3.1+cu121
  
• Python == 3.10.12
  
• NumPy == 1.26.4
  
• pandas == 2.2.2
  
• scikit-learn == 1.5.2
  
• SciPy == 1.13.1

Data Preparation

1. Download the following EEG Datasets:
   DEAP
   SEED
   
2. Unzip the data and organize it according to the following directory structure:
   

    data
    +---DEAP
    ¦   ¦   sXX.dat
    ¦   +---preprocessed
    ¦           Data_Orig_sub_XX.npy
    ¦           DE_sub_XX.npy
    ¦           labels_sub_XX.npy
    ¦           PSD_sub_XX.npy       
    +---SEED
        ¦   label.mat
        ¦   X_1.mat
        ¦   X_2.mat
        ¦   X_3.mat 
        +---preprocessed
                Data_Orig_sub_X.npy
                DE_sub_X.npy
                labels_sub_X.npy
                PSD_sub_X.npy

The preprocessed directory stores the data generated during the preprocessing stage in the code.

Execute Code

Run main.py to reproduce the results. This script will automatically link and execute the other .py files as needed.

Citation

If you find this code useful, please consider citing our work:

@article{imtiaz2025enhanced,
  title={Enhanced cross-dataset electroencephalogram-based emotion recognition using unsupervised domain adaptation},
  author={Imtiaz, Md Niaz and Khan, Naimul},
  journal={Computers in Biology and Medicine},
  volume={184},
  pages={109394},
  year={2025},
  publisher={Elsevier}
}