Table 5.

A comparison with the classification results from the literature on the vibration bearing dataset of the first case study.

Ref	Method	Testing Accuracy	Testing Time
[33]	Raw vibration with entropic features + SVM	98.9 ± 1.2	_
	Compressed sampled with α = 0.5 followed by signal reconstruction + SVM	92.4 ± 0.5
	Compressed sampled with α = 0.25 followed by signal reconstruction + SVM	84.6 ± 0.41
[44]	GP generated feature sets (un-normalised data)
	ANN	96.5
	SVM	97.1
[45]	FMM-RF                      SamEn	99.7 ± 0.02	_
	PS	99.7 ± 0.50
	SamEn + PS	99.8 ± 0.41
[36]	CPDC (with 6000 inputs from FFT)	99.4 ± 0.5	64.9
	CS-CPDC α = 0.1	99.8 ± 0.2	6.7
	α = 0.2	99.9 ± 0.1	7.8
[37]	With FFT, α = 0.1, feature dimension = 120, and LRC classifier)		_
	CS-FS	99.7 ± 0.4
	CS-LS	99.5 ± 0.3
	CS-Relief-F	99.8 ± 0.2
	CS-PCC	99.8 ± 0.3
	CS-Chi-2	99.5 ± 0.5
[46]	Feature selection (with λ = 0.004, tolerance value = 0.02) from compressively sampled data and SVM for fault classification:		_
	α = 0.1 and feature dimension = 14	98.8 ± 2.4
	α = 0.2 and feature dimension = 13	99.9 ± 0.2
	α = 0.3 and feature dimension = 26	99.9 ± 0.1
	Our proposed method with λ = 0.003, NCA tolerance value = 0.01, α = 0.1, and feature dimension = 8:
	MLR classifier	99.5 ± 0.6	0.015
	SVM classifier	99.5 ± 0.5	0.060
	Our proposed method with λ = 0.003, NCA tolerance value = 0.01, α = 0.2, and feature dimension = 10:
	MLR classifier	99.7 ± 0.3	0.003
	SVM classifier	99.8 ± 0.2	0.040
	Our proposed method with λ = 0.003, NCA tolerance value = 0.01, α = 0.3, feature dimension = 8:
	MLR classifier	100 ± 0.0	0.003
	SVM classifier	100 ± 0.0	0.030