Table 7 Comparison of methodologies presented in the literature for the five-class (Z-O-N-F-S) problem
From: Spectral information of EEG signals with respect to epilepsy classification
Authors | Feature extraction | Classification | Validation | Classification accuracy |
|---|---|---|---|---|
Guler and Ubeyli [32] (2005) | DWT (db2)/mean, min, max, std | Adaptive neuro-fuzzy inference system | 50% holdout | 98.68% |
Ubeyli and Guler [33] (2007) | Eigenvector methods (Pisarenko, MUSIC, Minimum-Norm) | Modified mixture-of-experts | 50% holdout | 98.60% |
Tzallas et al. [17] (2009) | TFD (SPWVD)/fractional energy | ANN | Monte Carlo cross-validation (50% split—10 repeats) | 89% |
Liang et al. [19] (2010) | FFT/ApEn | SVM | Monte Carlo cross-validation (60–40% split—10 repeats) | 85.90% |
Nicolaou et al. [34] (2012) | Permutation entropy | SVM | Monte Carlo cross-validation (various splits—100 repeats) | 86.10% |
Murugavel and Ramakrishnan [10] (2014) | DWT (db2)/energy, entropy, mean, min, max, std | OELM | 50% holdout | 94% |
Tawfik et al. [35] (2016) | Weighted permutation entropy | SVM | 10-fold cross-validation | 93.75% |
This study | Frequency sub-bands/energy, total energy, fractional energy, entropy | Random forests | 10-fold cross-validation | 91.20% |