Table 2.
Benchmarking of the previous seizure-prediction methods and our MViT approach: Kaggle/AES Seizure Prediction dataset.
| Authors/ Team |
Year | EEG Features | Classifier | SENS (%) |
AUC Score Public/Private |
|---|---|---|---|---|---|
| Medrr [32] | 2016 | N/A | N/A | - | 0.903/0.840 |
| QMSDP [32] | 2016 | Correlation, Hurst exponent, | LassoGLM, | - | 0.859/0.820 |
| fractal dimensions, | Bagged SVM, | ||||
| Spectral entropy | Random Forest | ||||
| Birchwood [32] | 2016 | Covariance, spectral power | SVM | - | 0.839/0.801 |
| ESAI CEU-UCH [32] | 2016 | Spectral power, | Neural Network, | - | 0.825/0.793 |
| correlation, PCA | kNN | ||||
| Michael Hills [32] | 2016 | Spectral power, correlation, | SVM | - | 0.862/0.793 |
| spectral entropy, fractal dimensions | |||||
| Truong et al. [24] | 2018 | EEG Spectrogram | CNN | 75.0 | - |
| Eberlein et al. [56] | 2018 | Multi-channel time series | CNN | - | 0.843/- |
| Ma et al. [57] | 2018 | Spectral power, correlation | LSTM | - | 0.894/- |
| Korshunova et al. [58] | 2018 | Spectral power | CNN | - | 0.780/0.760 |
| Liu et al. [27] | 2019 | PCA, spectral power | Multi-view CNN | - | 0.837/0.842 |
| Qi et al. [28] | 2019 | Spectral power, variance, correlation | Multi-scale CNN | - | 0.829/0.774 |
| Chen et al. [59] | 2021 | EEG Spectrogram | CNN | 82.00 | 0.746/- |
| Hussein et al. [29] | 2021 | EEG Scalogram | SDCN | 88.45 | 0.928/0.856 |
| Usman et al. [60] | 2021 | statistical and spectral moments | Ensemble of SVM, | 94.20 | - |
| CNN, and LSTM | |||||
| Zhao et al. [61] | 2022 | Raw EEG | CNN | 91.77–93.48 | 0.953–0.977/- |
| Proposed Method | 2022 | EEG Scalogram | MViT | 90.28 | 0.940/0.885 |