Table 4.

Performance of the proposed algorithm under different values of $S$.

Dataset	Metric	4	5	6	7	8
Tawsifur	Noise number	41	38	36	31	29
	RE (%)	78.87	80.41	81.44	84.02	85.05
	OA (%)	97.86	98.07	98.14	98.43	98.57
Skytells	Noise number	26	19	16	16	14
	RE (%)	87.91	91.16	92.56	92.56	93.49
	OA (%)	97.84	98.62	98.75	98.75	99.48
CXRs	Noise number	37	29	34	31	27
	RE (%)	78.49	83.14	80.23	81.98	84.30
	OA (%)	85.14	90.69	89.63	89.88	92.82

Note: bold numbers represent the best results. Experimental results of NLA are as follows: Noise number = 194, OA = 71.36 % (Tawsifur dataset); Noise number = 215, OA = 68.66 % (Skytells dataset); Noise number = 172, OA = 60.82 % (CXRs dataset).