A Multiattention-Based Supervised Feature Selection Method for Multivariate Time Series

Table 2

Regression and classification accuracy with different percentages of selected features with the RF classifier.

	SRU	DC	AEP	EEG (%)	OD (%)	WFRN (%)
Top 5% of selected features
m-AFS	1.65	6.55	3.09	89.98	95.38	92.74
Fisher; ’s score	2.63	11.24	3.72	81.35	98.52	93.47
ReliefF	2.74	9.39	5.53	70.09	97.48	94.30
Trace	2.80	5.94	4.05	68.09	84.74	90.71
LS_l21	1.86	6.81	3.36	63.81	99.01	92.46
RF	1.40	5.37	3.29	82.15	98.58	98.34
Top 10% of selected features
m-AFS	1.3	3.18	3.22	95.16	99.26	95.96
Fisher; ’s score	2.55	7.31	3.38	85.63	98.65	96.04
ReliefF	2.52	7.73	5.29	76.54	98.46	95.31
Trace	2.58	5.62	3.44	75.67	86.77	92.09
LS_l21	1.57	5.24	3.23	67.65	98.52	93.29
RF	1.10	3.88	3.61	85.36	99.20	98.07
Top 15% of selected features
m-AFS	0.99	2.80	3.27	96.26	99.26	95.96
Fisher; ’s score	2.52	7.03	3.30	89.31	98.71	96.87
ReliefF	2.51	6.14	4.73	82.73	98.77	95.59
Trace	2.54	5.40	3.15	79.89	86.95	92.56
LS_l21	1.50	4.56	3.20	75.01	98.52	93.38
RF	0.96	2.81	3.60	86.97	99.32	98.07