Improving Predictions of Multiple Binary Models in ILP

Table 4

Average one-versus-rest AUCs of our multi-class methods (MRL, MRSU, and MRSI) compared to CN2, with average ranks in brackets. The AUCs reported for Aleph are for reference only, as these arise from overemphasising the default rules.

	MRL	MRSU	MRSI	CN2	Aleph standard
Data set 1	83.03  (1.00)	75.2  (3.00)	73.92  (4.00)	75.39  (2.00)	82.80
Data set 2	88.72 (4.00)	89.65  (1.00)	88.90  (3.00)	89.58  (2.00)	88.66
Data set 3	91.97 (3.00)	92.63 (2.00)	93.38  (1.00)	91.43  (4.00)	86.78
Data set 4	70.50 (4.00)	72.67 (2.00)	74.15  (1.00)	72.60  (3.00)	66.33
Data set 5	87.28 (2.00)	86.62 (4.00)	89.03  (1.00)	86.63  (3.00)	83.29
Data set 6	82.05  (1.00)	74.03  (3.00)	81.41  (2.00)	73.27  (4.00)	76.38
Average	83.92 (2.50)	81.80 (2.50)	83.46  (2.00)	81.48  (3.00)	80.71
Data set 7	64.03  (1.00)	57.19  (4.00)	57.28  (2.00)	57.19  (3.00)	63.93
Data set 8	60.77  (1.00)	51.70  (3.00)	57.10  (2.00)	51.39  (4.00)	64.03
Data set 9	74.48  (1.00)	63.91  (3.00)	72.38  (2.00)	60.93  (4.00)	60.90
Data set 10	55.07  (1.00)	52.70  (3.50)	52.70  (2.00)	52.70  (3.50)	55.07
Data set 11	65.46  (1.00)	56.15  (3.00)	57.06  (2.00)	55.53  (4.00)	64.71
Average	63.96  (1.00)	56.33  (3.30)	59.30  (2.00)	55.55  (3.70)	61.73