Epileptic Seizure Detection Using a Hybrid 1D CNN-Machine Learning Approach from EEG Data

Table 6

Comparison of previous studies conducted for epilepsy detection.

Study	Features extraction	Classification method	Datasets	Classes	Accuracy (%)
[1]	Chebyshev IIR filter, discrete wavelet transform	SVM	Bonn	2	96
		ANN			98
[3]	None	CNN	Bonn	2	99.52
				3	96.97
				5	93.55
[4]	None	CNN	Bonn	3	88.67
[6]	Empirical mode decomposition (EMD), intrinsic mode function (IMF)	Classification and regression tree (CART)	Bonn	3	93.55
[7]	Recurrence quantification analysis (RQA)	SVM	Bonn	3	95.60
[8]	Channel selection and statistical feature extraction	Ensemble	CHB-MIT	2	89.02%
[9]	Tunable-Q wavelet transform (TQWT)	Random Forest (RF)	Bonn	3	99
[10]	Multiscale PCA, wavelet packet decomposition	SVM	Bonn	3	99.70
[11]	Discrete wavelet transform, temporal and spectral features	Fuzzy rough	CHB-MIT	2	92.79%
		Nearest neighbor
[12]	None	1D-pyramidal CNN	Bonn	3	99.1
[13]	None	1D-feature fusion CNN	Bonn	3	98.67
[14]	CWT	CNN	Bonn	2	100
				3	99
				4	91.50
				5	93.60
[44]	Time-frequency analysis (TFA)	ANN	Bonn	2	100
				5	89
[45]	Symplectic geometry eigenvalues	SVM	CHB-MIT	2	99.62
[46]	Adaptive-rate FIR filtering and DWT + MI-based feature selection	Ensemble of MLP, k-NN, SVM, BG, and RF	Bonn	2	100
		—		3	99.50
				4	96
				5	92
			CHB-MIT	2	99.38
Our approach	CNN	ML classifiers	Bonn	2	100
				3	99.33
				4	96
				5	94
	—	—	CHB-MIT	2	97.1