Using natural language processing methods to classify use status of dietary supplements in clinical notes

Table 3 Performances of five classification algorithms with different feature sets in the test data

Type	Features	Decision tree			Random forest			Naïve Bayes			SVM			Maximum Entropy
Type	Features	P	R	F	P	R	F	P	R	F	P	R	F	P	R	F
Type 1	raw uni^a	0.819	0.817	0.816	0.858	0.853	0.853	0.770	0.757	0.755	0.818	0.816	0.815	0.850	0.849	0.849
Type 2	uni	0.846	0.845	0.844	0.878	0.876	0.876	0.793	0.784	0.783	0.837	0.835	0.834	0.874	0.873	0.873
Type 3	tf-idf	0.862	0.857	0.857	0.862	0.857	0.857	0.763	0.704	0.701	0.844	0.839	0.839	0.840	0.831	0.831
Type 4	bi^a	0.760	0.720	0.716	0.760	0.720	0.716	0.715	0.707	0.702	0.735	0.719	0.720	0.749	0.739	0.739
Type 5	uni + bi	0.872	0.864	0.863	0.872	0.864	0.863	0.815	0.808	0.807	0.881	0.877	0.876	0.890	0.888	0.887
Type 6	uni + bi+tri^a	0.863	0.852	0.850	0.863	0.852	0.850	0.815	0.808	0.808	0.880	0.876	0.875	0.887	0.883	0.882
Type 7	indi^a only	0.848	0.847	0.846	0.861	0.860	0.860	0.860	0.849	0.848	0.851	0.849	0.849	0.862	0.859	0.859
Type 8	uni + bi+indi	0.860	0.860	0.860	0.875	0.865	0.864	0.813	0.803	0.801	0.899	0.897	0.897	0.895	0.903	0.902
Type 9	uni + bi+tri + indi	0.860	0.857	0.857	0.872	0.861	0.860	0.813	0.803	0.801	0.899	0.897	0.897	0.905	0.903	0.902

ISSN: 1472-6947