A Novel Approach to Ensemble Classifiers: FsBoost-Based Subspace Method

Table 7

Level 1 classification results for the D dataset.

k-nearest neighbor classification algorithm
NP	k = 2			k = 1			k = 5
	DF
	Euclidean			Minkowski			Euclidean
FS	0			1			2
NF	68			25			15
Class	Sen	Spe	Acc	Sen	Spe	Acc	Sen	Spe	Acc
E	0.79	0.97	88.04	0.82	0.84	89.17	0.84	0.92	88.13
N-E	0.97	0.79		0.97	0.92		0.92	0.84
AUC	0.88			0.78			0.90
Kappa	0.76			0.89			0.80
F-measure	0.87			89.17			0.89
10-fold (%)	84.24			86.30			87.17
Probabilistic neural networks
NP	Spread			Spread			Spread
	0.41			0.41			0.41
FS	0			1			2
NF	68			25			15
Class	Sen	Spe	Acc	Sen	Spe	Acc	Sen	Spe	Acc
E	0.68	0.97	82.43	0.64	0.53	79.52	0.53	0.95	74.09
N-E	0.97	0.68		0.95	0.95		0.95	0.53
AUC	0.82			0.59			0.80
Kappa	0.65			0.76			0.60
F-measure	0.80			79.52			0.77
10-fold (%)	0.41			0.41			0.41
Support vector machines
NP	BoxConstraint
	1			21			2
FS	0			1			2
NF	68			25			15
Class	Sen	Spe	Acc	Sen	Spe	Acc	Sen	Spe	Acc
E	0.97	0.93	94.61	0.93	0.93	93.78	0.93	0.92	92.30
N-E	0.93	0.97		0.94	0.92		0.92	0.93
AUC	0.95			0.88			0.94
Kappa	0.89			0.94			0.88
F-measure	0.95			93.78			0.94
10-fold (%)	95.43			94.54			92.33

DF: distance function, Sen: sensitivity, Spe: specificity, Acc: accuracy (%), NP: network parameters, FS: feature selection, NF: number of features, EC: ensemble classifier, E: epilepsy, and N-E: nonepilepsy.