A DM-ELM based classifier for EEG brain signal classification for epileptic seizure detection

Figures & data

Figure 1. Sample EEG signal from each set A, B, C, D and E.

Figure 2. Sample EEG signal from each set A, B, C and D after decomposition using DWT.

Figure 3. Sample EEG signal from set E after decomposition using DWT.

Figure 4. Architecture of proposed model (DMELM).

Figure 5. Accuracy comparison of different Afs for SET 1, 2, and 3 using ELM.

Table 1. Parameter setup for different techniques used.

Techniques used	Parameter	Value
Moth Flame Optimization	Max_Iteration	1000
	Lower Bound (lb)	100
	Upper Bound (ub)	100
	Number of search agents	50
	Dimension of search space	30
	Objective function	$f_H\left(x\right)=\sum _{i=1}^H\sum _{j=1}^N{v}_{i}h\left(w_i\ast x_j+{\beta }_i\right)$
Extreme Learning Machine	Activation function	Inverse Multiquadric
	Number of hidden layers	500

Table 2. Accuracy comparison for different activation functions for SET 1, 2 and 3 using ELM.

ELM Activation Functions	Dataset
	SET-1	SET-2	SET-3
tanh	0.71 (± 0.11)	0.82 (± 0.17)	0.70 (± 0.09)
sine	0.82 (± 0.10)	0.94 (± 0.06)	0.82 (± 0.09)
tribas	0.49 (± 0.45)	0.50 (± 0.45)	0.50 (± 0.45)
inv_tribas	0.11 (± 0.21)	0.10 (± 0.20)	0.10 (± 0.20)
sigmoid	0.76 (± 0.10)	0.87 (± 0.11)	0.74 (± 0.10)
hardlim	0.11 (± 0.21)	0.10 (± 0.20)	0.10 (± 0.20)
softlim	0.11 (± 0.21)	0.10 (± 0.20)	0.10 (± 0.20)
gaussian	0.59 (± 0.22)	0.34 (± 0.32)	0.59 (± 0.21)
multiquadric	0.88 (± 0.07)	0.89 (± 0.13)	0.89 (± 0.06)
inv_multiquadric	0.86 (± 0.11)	0.87 (± 0.12)	0.87 (± 0.10)

Figure 6. Validation curve for EEG signal classification using DM- Extreme Learning Machine (ELM) with three different categories of dataset (Set1, Set2 and Set3).

Table 3. Performance comparison for different machine learning techniques for SET1.

Classification techniques	Accuracy	Specificity	Sensitivity	F-score
MLPNN	0.84 ± 0.17	0.81 ± 0.12	0.82 ± 0.11	0.82 ± 0.10
Naïve Bayesian	0.78 ± 0.13	0.79 ± 0.15	0.76 ± 0.03	0.78 ± 0.15
ELM	0.88 ± 0.07	0.85 ± 0.05	0.84 ± 0.06	0.82 ± 0.08
DM-ELM	0.92 ± 0.02	0.93 ± 0.12	0.91 ± 0.03	0.94 ± 0.13

Table 4. Performance comparison for different machine learning techniques for SET2.

Classification techniques	Accuracy	Specificity	Sensitivity	F-score
MLPNN	0.85 ± 0.11	0.84 ± 0.01	0.84 ± 0.14	0.85 ± 0.13
Naïve Bayesian	0.79 ± 0.09	0.78 ± 0.12	0.77 ± 0.07	0.77 ± 0.07
ELM	0.89 ± 0.13	0.88 ± 0.14	0.87 ± 0.12	0.90 ± 0.04
DM-ELM	0.95 ± 0.03	0.94 ± 0.13	0.96 ± 0.02	0.96 ± 0.13

Table 5. Performance comparison for different machine learning techniques for SET3.

Classification techniques	Accuracy	Specificity	Sensitivity	F-score
MLPNN	0.84 ± 0.14	0.82 ± 0.11	0.83 ± 0.17	0.84 ± 0.04
Naïve Bayesian	0.80 ± 0.12	0.79 ± 0.14	0.78 ± 0.13	0.79 ± 0.16
ELM	0.89 ± 0.06	0.88 ± 0.05	0.87 ± 0.03	0.88 ± 0.08
DM-ELM	0.96 ± 0.12	0.95 ± 0.13	0.97 ± 0.02	0.94 ± 0.03