From: Generative adversarial networks in EEG analysis: an overview
Study | Purpose | Dataset | GAN type | Evaluation metrics | Results (with GAN) |
---|---|---|---|---|---|
Y. Luo et al., 2018 [56] | Enhance EEG-based emotion recognition | SEED + DEAP | cWGAN | Classification accuracy | SEED: + 2.97% DEAP-Arousal: + 9.15% DEAP-Valence: + 20.13% |
Y. Luo et al., 2018 [57] | Enhance EEG-based emotion recognition for semi-supervised models | SEED + DEAP | WGANDA | Classification accuracy | (W.R.T: SVM) SEED: + 30.43% DEAP-Arousal: + 17.63% DEAP-Valence: + 17.63% |
Y. Luo et al., 2020 [59] | Enhance EEG-based emotion recognition | SEED + DEAP | cWGAN  +  sWGAN | Classification accuracy | SEED cWGAN + DNN: + 8.3% sWGAN + DNN: + 10.2% DEAP cWGAN + SVM: + 3.5% sWGAN + SVM: + 5.4% |
Dong and Ren, 2020 [60] | Enhance EEG-based emotion recognition | DEAP | MCLFS-GAN | Classification accuracy | (w.r.t CNN + LSTM) SAP MCLFS-GAN: + 14.95% LOSO MCLFS-GAN: + 19.52% |
Fu et al., 2021 [62] | Achieve a fine mapping of EEG data directly to facial images | SEED | Ac-GAN | • Classification accuracy • Reliability • Entropy | • Generated images from EEG 82.14% • 92.02% • 7.41 |
Liang et al. 2021 [63] | Fuse the spatial and temporal dynamic brain information into a better feature representation | SEED + DEAP + MAHNOB-HCI | CNN + RNN + GAN | • Classification accuracy • F1 score | • Up to + 7.69% • Up to + 5.07 |
Pan and Zheng 2021 [64] | Enhance EEG-based emotion recognition with sample scarcity and category imbalance issues | DEAP + MAHNOB-HCI | PSD-GAN | Recognition accuracy | • 2-classification task: 5.25%: 6.71% • 4-classification task: 10.92%: 14.47% |
Chang and Jun 2019 [65] | Recognize the emotional responses of users towards given architectural design | Chang, Dong, and Jun Dataset [66] | GAN | Classification accuracy |  + 0.5% |