Table 3.

Obesity: Comparison of F1 measure for obesity models between 2008, 2014 and combined 2008/2014 broken down by SVM kernel, weighting scheme, and feature space.

Kernel	Weight	Feature Space	2008	2014	Combined 2008/2014	% Change 2008, 2014	% Change 2008, Comb.	% Change 2014, Comb.
RBF	tf-idf	Min, Lex & CUI	0.930	0.945	0.945	1.65	1.65	0.00
RBF	tf-idf	Min	0.935	0.939	0.945	0.48	1.10	0.62
RBF	tf-idf	Lex	0.935	0.933	0.945	−0.16	1.10	1.26
RBF	Count	Lex	0.940	0.945	0.945	0.55	0.55	0.00
RBF	tf-idf	Min & Lex	0.940	0.945	0.940	0.55	0.00	−0.55
RBF	Count	Min, Lex & CUI	0.915††	0.945	0.940	3.30**	2.73*	−0.55
RBF	Count	Min & Lex	0.891††	0.945	0.940	6.04**	5.46**	−0.55
RBF	Count	Min & CUI	0.925	0.945	0.940	2.20	1.64	−0.55
RBF	Count	Min	0.940	0.921	0.940	−1.97	0.00	2.01
RBF	Count	Lex & CUI	0.935	0.940	0.940	0.55	0.55	0.00
Linear	tf-idf	Min & CUI	0.934	0.871††	0.940	−6.73**	0.62	7.89**
RBF	tf-idf	Lex & CUI	0.940	0.945	0.939	0.55	−0.07	−0.62
RBF	tf-idf	Min & CUI	0.915††	0.907††	0.939	−0.87	2.66	3.56*
Linear	Count	Min & CUI	0.923	0.892††	0.933	−3.41	1.11	4.68
Linear	Count	Min	0.923	0.840††	0.933	−9.05**	1.11	11.18**
Linear	Count	Lex	0.928	0.912†	0.933	−1.71	0.56	2.31
Linear	Count	Min, Lex & CUI	0.923	0.898†	0.928	−2.69	0.55	3.34
RBF	tf-idf	CUI	0.928	0.886††	0.928	−4.49**	0.00	4.70**
Linear	tf-idf	Lex	0.929	0.913††	0.926	−1.69	−0.35	1.36
Linear	Count	Min & Lex	0.909††	0.940	0.922	3.39**	1.44	−1.88
Linear	tf-idf	Min	0.940	0.906††	0.921	−3.62*	−1.97	1.71
Linear	tf-idf	Min, Lex & CUI	0.899††	0.906††	0.920†	0.78	2.40	1.61
Linear	tf-idf	Min & Lex	0.940	0.913††	0.905††	−2.83	−3.71**	−0.91
Linear	Count	Lex & CUI	0.945	0.892††	0.903††	−5.66**	−4.47**	1.27
Linear	tf-idf	Lex & CUI	0.939	0.878††	0.894††	−6.51**	−4.80**	1.83
Linear	tf-idf	CUI	0.867††	0.846††	0.821††	−2.42	−5.26*	††2.92
RBF	Count	CUI	0.820††	0.385††	0.755††	−53.11**	−7.96	96.29**
Linear	Count	CUI	0.693††	0.316††	0.370††	−54.44**	−46.57**	17.28

^*percent change significant at alpha = 0.05

^**percent change significant at alpha = 0.1

^†statistically significantly different from highest performing model (emphasized above) at alpha = 0.05

^††statistically significantly different from highest performing model (emphasized above) at alpha =0.1

Note: For visual reference, models which are not statistically significantly different from the emphasized model at alpha = 0.1 are shaded in dark grey. Models which are not statistically significantly different from the emphasized model at alpha = 0.05 are shaded in light grey.