Table 2.

System performance, and results of support vector machine (SVM) retraining using 10-fold validation

		SVM retraining results
Model	Overall system accuracy (95% CI)	Class	Precision (95% CI)	Recall (95% CI)	F-measure (95% CI)
Original cTAKES, no changes, polynomial SVM	0.69 (0.65 to 0.72)	C	0.3 (0.24 to 0.35)	0.81 (0.72 to 0.89)	0.43 (0.36 to 0.51)
		P	0.87 (0.81 to 0.91)	0.6 (0.54 to 0.66)	0.71 (0.64 to 0.77)
Modified cTAKES rules, original keywords, polynomial SVM	0.9 (0.87 to 0.92)	C	0.66 (0.57 to 0.75)	0.77 (0.68 to 0.85)	0.71 (0.62 to 0.8)
		P	0.92 (0.88 to 0.95)	0.83 (0.78 to 0.88)	0.87 (0.83 to 0.91)
Modified rules, grouped keywords, polynomial SVM default weights	0.92 (0.9 to 0.94)	C	0.75 (0.66 to 0.82)	0.89 (0.81 to 0.94)	0.81 (0.72 to 0.88)
		P	0.97 (0.93 to 0.99)	0.87 (0.82 to 0.91)	0.91 (0.87 to 0.95)
Modified rules, grouped keywords, RBF SVM, default weights	0.9 (0.88 to 0.92)	C	0.81 (0.69 to 0.89)	0.56 (0.45 to 0.66)	0.66 (0.55 to 0.76)
		P	0.85 (0.81 to 0.89)	0.93 (0.89 to 0.96)	0.89 (0.85 to 0.92)
Modified rules, grouped keywords, sigmoid SVM, default weights	0.84 (0.81 to 0.87)	C	0.49 (0.39 to 0.59)	0.53 (0.42 to 0.63)	0.51 (0.4 to 0.61)
		P	0.82 (0.77 to 0.87)	0.77 (0.72 to 0.82)	0.8 (0.74 to 0.84)

C, current smoker; cTAKES, clinical Text Analysis and Knowledge Extraction System; P, past smoker; RBF, radial basis function.