Knowledge Mining from Clinical Datasets Using Rough Sets and Backpropagation Neural Network

Table 11

Accuracy of reducts for (a) hepatitis, (b) breast cancer, and (c) heart disease.

(a) Hepatitis accuracy (%)

Reduct	(80–20)	(70–30)	(60–40)
	95.2	93.9	91.2
	96.6	92.5	93.2
	93.9	91.2	92.5
	95.2	91.8	91.2
	93.9	91.8	94.6
	95.9	94.6	91.8
	95.2	93.2	91.2
	95.2	92.5	91.8
	97.3	95.9	93.9
All features	87.8	87.1	86.4

(b) Breast cancer accuracy (%)

Reduct	(80–20)	(70–30)	(60–40)
	98.6	98.3	97.6
	97.6	97.4	97.4
	98.4	98.0	97.6
	98.4	98.1	97.4
	98.0	97.9	97.6
	98.0	98.0	97.9
	97.9	97.7	97.1
	98.4	98.0	97.3
	98.0	97.7	97.6
All features	97.7	97.7	97.4

(c) Heart Disease Accuracy

Reduct	(80–20)	(70–30)	(60–40)
	79.6	78.1	77.4
	81.1	81.1	80.4
	84.8	84.8	84.4
	85.9	83.0	83.3
	88.9	84.8	84.1
	90.4	88.1	87.8
	88.1	87.0	85.9
	86.3	85.2	83.3
	88.5	85.6	84.8
All features	84.4	84.1	84.8