Improving sensitivity of machine learning methods for automated case identification from free-text electronic medical records

Table 4 Sensitivity and specificity of various classifiers trained on the acute renal failure data set for difference percentages of under-sampling

Under-sampling	SVM		MyC		RIPPER		C4.5		Imbalance
(%)	Sens.	Spec.	Sens.	Spec.	Sens.	Spec.	Sens.	Spec.	ratio
0	0.62	0.92	0.69	0.90	0.71	0.89	0.69	0.88	16
10	0.64	0.90	0.74	0.89	0.75	0.89	0.69	0.87	14
20	0.64	0.89	0.75	0.83	0.75	0.88	0.74	0.86	13
30	0.66	0.88	0.76	0.82	0.76	0.88	0.75	0.85	11
40	0.70	0.85	0.75	0.87	0.74	0.88	0.75	0.85	9
50	0.74	0.81	0.76	0.80	0.77	0.76	0.76	0.82	8
60	0.82	0.72	0.77	0.81	0.84	0.68	0.83	0.82	6
70	0.83	0.67	0.83	0.70	0.83	0.61	0.86	0.77	5
80	0.86	0.56	0.89	0.49	0.90	0.44	0.90	0.45	3
90	0.92	0.41	0.90	0.43	0.89	0.43	0.92	0.39	2

ISSN: 1472-6947