Perceptron Ranking Using Interval Labels with Ramp Loss for Online Ordinal Regression

Table 4

Sensitivity estimation of PRIL, PA-I, and proposed PRIL-RAMP algorithm of three datasets under 50% noise data.

Dataset	Rounds	Method	MAE	RMSE	Spearman’s correlation coefficient	Discard sample	Discard rate	Time (s)
Abalone	T = 1400	PA-I-50%					0.00	0.296
		PRIL-50%					0.00	0.217
		PRIL-RAMP-50%					30.21	0.817
	T = 2800	PA-I-50%					0.00	0.558
		PRIL-50%					0.00	0.425
		PRIL-RAMP-50%					34.14	0.369
	T = 4100	PA-I-50%					0.00	0.805
		PRIL-50%					0.00	0.597
		PRIL-RAMP-50%					42.51	0.551
Parkinsons-uprds	T = 2000	PA-I-50%					0.00	0.094
		PRIL-50%					0.00	0.345
		PRIL-RAMP-50%					47.10	0.333
	T = 4000	PA-I-50%					0.00	0.128
		PRIL-50%					0.00	0.670
		PRIL-RAMP-50%					46.90	0.628
	T = 5800	PA-I-50%					0.00	0.273
		PRIL-50%					0.00	0.978
		PRIL-RAMP-50%					47.32	0.994
Real estate valuation	T = 140	PA-I-50%					0.00	0.159
		PRIL-50%					0.00	0.168
		PRIL-RAMP-50%					32.85	0.178
	T = 280	PA-I-50%					0.00	0.324
		PRIL-50%					0.00	0.346
		PRIL-RAMP-50%					33.21	0.362
	T = 400	PA-I-50%					0.00	0.475
		PRIL-50%					0.00	0.517
		PRIL-RAMP-50%					41.75	0.537