Table 3 Overview of existing work on seizure detection using—machine learning classifiers, features, performance score, performance metrics, datasets, and Authors
From: A review of epileptic seizure detection using machine learning classifiers
Classifier(s) | Feature(s) | Performance (%) | Performance metrics | Dataset | Authors |
|---|---|---|---|---|---|
SVM | Vector | 96 | Sensitivity (Sen) | CHB-MIT | Shoeb and Guttag [41] |
Random forest | Time and frequency | 93.8 | Senstivity | EPILEPSIAE | Donos et al. [44] |
ANN | Line length | 99.6 | Classification accuracy (Class Acc) | BONN | Guo et al. [69] |
Burst detection algo | Line length | 84.27, 84,85.7 | Acc, Sen, Specificity (Spec) | NICU, Belgium | Koolen et al. [70] |
Normalization | Line length | 52 | ROC | CHB-MIT | Logesparan et al. [71] |
ELM and BPNN | SE | 95.6 | Class Accuracy | BONN | Song and Lio [72] |
SVM and ELM | AE and SE | 95.58 | Class Accuracy | BCI Lab, Colarodo | Zhang et al. [73] |
SVM | DWT | 94.8 | Avg Accuracy | CHB-MIT | Ahmad et al. [74] |
GMM | Spectral, hybrid, temporal | 87.58 | Avg Accuracy | CHB-MIT | Gill et al. [75] |
Random forest | PCA, STF, Moving Max | 97.12, 99.29, 0.77/h | Sen, Spec, FPR | CHB-MIT | Orellana and Cerqueira [76] |
Random forest and k-NN | Spectral power | 80.87, 47.45, 2.5/h, 56.23 | Sen, Prec, FPR, F-meas | CHB-MIT | Birjandtalab et al. [77] |
Boosting | Stockwell | 94.26, 96.34 | Sen, Spec | Freiburg | Yan et al. [78] |
SVM, MLP, KNN, Naïve bayes | Energy | 98.75 | Class Acc | EPILEPSIAE | Amin et al. [79] |
Random forest | Entropy and DWT | 98.45 | Class Acc | BONN | Mursalin et al. [80] |
SVM | Time–Frequency | 90.62, 99.32 | Sen, Spec | CHB-MIT | Zabihi et al. [81] |
Random forest | Time-domain | 96.94 | ROC curve | Kaggle | Truong et al. [82] |
SVM, LDA, QDA, LC,PC, DT, KNN, UDC, PARZEN | Time–frequency | 84, 85 | Sen, Spec | CHB-MIT | Fergus et al. [83] |
SVM | DWT | 86.83 | Confusion Matrix | CHB-MIT | Chen et al. [84] |
SVM and neural network | DWT and CWT | 99.1 | Overall Acc | BONN | Satapathy et al. [85] |
ELM | Time–frequency | 97.73, 0.37/h | Sen, false alarm rate | Freiburg | Yuan et al. [86] |
SVM | DWT | 99.38 | Class Acc | BONN | Subasi et al. [87] |
LS-SVM | FFT and DWT | 100 | Class Acc | BONN | Al Ghayab et al. [88] |
SVM and Naïve bayes | Entropy, RMS, variance, energy | 96.55, 95.63, 95.7 | Sen, Spec, Acc | CHB-MIT | Selvakumari et al. [89] |
LS-SVM | 8 types of Entropies | 100, 99.4, 99.5 | Sen, Spec, Acc | BONN | Chen S et al. [90] |
ANN | Spectral power | 86 | F-meas | CHB-MIT | Birjandtalab et al. [91] |
KNN and GHE | - | 100 | Class Acc | BONN | Lahmiri and shumel [92] |
Random forest | DWT | 99.74, 0.21/h | Sen, FPR | BONN and Freiburg | Tzimourta et al. [93] |
Random forest | STFT, mean, energy and std dev | 96.7 | Class Acc | BONN | Wang et al. [94] |
Random forest, SVM, KNN, and Adaboost | 28 statistical and time–frequency features | 97.6, 94.4, 96.1, 92.9, 98.8, 0.96 | Sen, Spec, Acc, PPR, NPR, ROC | Bern-Barcelona | Raghu and Sriraam [95] |
ANN,KNN,SVM, and Random forest | Mean, std dev, power, skewness, kurtosis, absolute mean | 100 | Overall Accuracy | Freiburg and CHB-MIT | Alickovic et al. [96] |
SVM | Energy | 99.5 | Class Acc | BONN and Barcelona | Fasil and Rajesh [97] |
SVM and Random forest | 10-time and frequency | 0.98 | ROC(AUC) | EPILEPSIAE | Manzouri et al. [98] |
LS-SVM | DCT, SVD, IMF, DCT-DWT, | 91.36 | Acc, Sen, Spec | Freiburg | Parvez and Paul [99] |
SysFor and Forest CERN | 9 statistical features | 100 | Class Acc | Epilepsy Centre UCSF | Siddiqui et al. [63] |
Random forest | L1-penalized robust regression (L1PRR) | 100 | Class Acc | BONN | Hussein et al. [100] |
SVM, NB, KNN, random forest, logistic model Trees (LMT) | 15-features | 97.40, 97.40,97.50 | Acc, Sen, Spec | BONN | Mursalin et al. [101] |
Random forest | IMF | 98.4,98.6,96.4 | Sen, Spec, Acc | BONN | Sharma et al. [102] |
ANN | Time–frequency | 100 | Overall Acc | BONN | Tzallas et al. [103] |
Decision forest–Random forest, Boosting | 9 statistical features | 96.67,74.36, 84.06 | Pre, Rec, F-measure | CHB-MIT | Siddiqui et al. [104] |