. 2003;2003:440–444.

Table 1.

Model performance on cross validated data.

Abstract type	Training Cases^*		Model Type	Sentence Location Feature	Linear Classifier (Widrow-Huff)				SVM
Abstract type	Total	Positive	Model Type	Sentence Location Feature	ROC	Acc	P/R	F₁	ROC	Acc	P/R	F₁
Unstructreud	1,532	196 (12.8%)	Intro	N	0.863	0.889	0.71/0.34	0.465	0.910	0.921	0.82/0.49	0.616
		196 (12.8%)	Intro	Y	0.867	0.890	0.71/0.30	0.423	0.957	0.947	0.88/0.72	0.789
		430 (28.1%)	Method	N	0.851	0.829	0.85/0.48	0.611	0.935	0.894	0.87/0.74	0.800
		430 (28.1%)	Method	Y	0.854	0.832	0.86/0.48	0.612	0.954	0.909	0.84/0.84	0.837
		686 (44.8%)	Result	N	0.672	0.737	0.82/0.57	0.672	0.920	0.863	0.85/0.85	0.851
		686 (44.8%)	Result	Y	0.650	0.730	0.80/0.55	0.650	0.930	0.860	0.83/0.86	0.845
		220 (14.4%)	Concl	N	0.877	0.888	0.77/0.33	0.457	0.911	0.903	0.67/0.60	0.639
		220 (14.4%)	Concl	Y	0.883	0.891	0.77/0.33	0.457	0.965	0.936	0.81/0.74	0.773
Structured	1,669	314 (18.8%)	Intro	N	0.832	0.854	0.69/0.41	0.518	0.912	0.888	0.75/0.62	0.679
		Y	Intro	0.841	0.857	0.71/0.43	0.534	0.980	0.969	0.94/0.88	0.908
		547 (32.8%)	Method	N	0.758	0.764	0.76/0.40	0.524	0.910	0.867	0.80/0.78	0.790
		547 (32.8%)	Method	Y	0.754	0.755	0.75/0.39	0.511	0.909	0.858	0.81/0.75	0.778
		554 (33.2%)	Result	N	0.822	0.782	0.80/0.47	0.591	0.894	0.845	0.79/0.73	0.762
		554 (33.2%)	Result	Y	0.826	0.785	0.80/0.46	0.586	0.905	0.846	0.77/0.76	0.763
		249 (14.9%)	Concl	N	0.820	0.870	0.75/0.19	0.306	0.851	0.882	0.68/0.42	0.520
		249 (14.9%)	Concl	Y	0.823	0.868	0.70/0.20	0.307	0.974	0.954	0.86/0.82	0.840
	90,665	14,248 (15.7%)	Intro	N	0.876	0.890	0.80/0.41	0.545	0.933	0.924	0.80/0.92	0.746
		Y	Intro	0.873	0.890	0.79/0.41	0.541	0.975	0.967	0.92/0.97	0.892
		25,826 (28.5%)	Method	N	0.846	0.813	0.77/0.49	0.600	0.939	0.891	0.80/0.82	0.811
		25,826 (28.5%)	Method	Y	0.832	0.811	0.77/0.48	0.594	0.942	0.895	0.81/0.83	0.820
		34,671 (38.2%)	Result	N	0.831	0.786	0.78/0.61	0.687	0.929	0.871	0.81/0.86	0.835
		34,671 (38.2%)	Result	Y	0.816	0.783	0.78/0.60	0.678	0.922	0.860	0.81/0.83	0.821
		12,805 (14.1%)	Concl	N	0.880	0.893	0.73/0.36	0.478	0.939	0.918	0.74/0.63	0.682
		12,805 (14.1%)	Concl	Y	0.850	0.889	0.72/0.35	0.469	0.991	0.970	0.88/0.91	0.895

Data tested using 10 fold cross validation. For any given model 90% of all cases of numbers reported were used in model generation and 10% were used for holdout test set.