Table 3.

Number of features [# f.], precision [P.], recall [R.], F1 score [F1] of classifier top performances by dimensionality reduction technique.

	Naïve Bayes				SVM (linear kernel)				SVM (RBF kernel)				MaxEnt
	# f.	P.	R.	F1	# f.	P.	R.	F1	# f.	P.	R.	F1	# f.	P.	R.	F1
Full feature space	45,695	0.761	0.635	0.688	45,695	0.744	0.802	0.769	45,695	0.714	0.829	0.765	45,695	0.740	0.790	0.762
PCA	333	0.541	0.957	0.691	426	0.739	0.798	0.765	426	0.709	0.825	0.760	426	0.737	0.790	0.760
MI	228	0.628	0.821	0.710	2,284	0.746	0.799	0.770	22,847	0.728	0.825	0.772	11,423	0.753	0.775	0.761
Ontology	32,180	0.872	0.689	0.766	32,180	0.762	0.795	0.774	32,180	0.747	0.760	0.747	32,180	0.733	0.806	0.764
Ontology PCA	210	0.538	0.942	0.684	405	0.748	0.779	0.758	405	0.718	0.775	0.741	405	0.726	0.810	0.763
Ontology MI	24,135	0.879	0.693	0.771	1,605	0.747	0.798	0.769	1,605	0.715	0.795	0.747	8,027	0.723	0.790	0.752

Note: Performance of shaded region is the simplest model that is not significantly different from the top overall model (i.e., F1 of 77.4%).