Improved Feature-Selection Method Considering the Imbalance Problem in Text Categorization

Table 5

The comparison of nine improved and existing feature-selection methods with respect to micro-𝐹1 measure on Reuters-21578 for NB and SVM, respectively. The bold values indicate the best performance of the classifier when various feature-selection methods are used, respectively.

Feature selection	Naïve Bayes					Support vector machines
	400	800	1200	1600	2000	400	800	1200	1600	2000
IG	62.09	64.60	64.76	65.11	65.22	61.31	62.19	62.66	62.59	62.88
IGX	62.07	64.61	64.77	65.10	65.22	64.82	65.89	66.56	66.66	67.01
CHI	62.89	64.02	64.96	64.92	65.35	62.85	63.12	62.96	62.93	62.73
CHIX	64.66	65.70	66.00	66.07	65.94	67.36	67.29	67.28	67.13	66.82
MI	34.65	54.87	59.71	61.58	62.77	43.02	51.67	56.94	58.98	60.11
MIX	39.55	51.46	59.64	61.84	63.43	39.47	51.35	61.41	63.44	64.59
DF	62.41	64.09	64.99	65.36	65.51	61.07	62.75	62.69	62.70	62.82
DFX	63.99	65.38	65.65	65.92	65.94	65.90	66.65	67.05	67.01	67.58
GINI	65.13	65.82	66.64	66.16	65.78	63.42	63.17	63.07	62.95	62.71
GINIX	66.43	66.45	66.39	66.54	66.49	67.44	67.18	66.99	67.11	67.20
DIA	30.10	30.82	31.49	32.85	37.60	45.06	48.87	51.05	51.85	53.04
DIAX	48.11	57.71	63.69	64.23	64.95	49.79	59.15	65.73	66.39	66.96
CMFS	66.03	66.52	66.38	66.66	66.64	63.73	63.52	63.14	63.06	62.97
CMFSX	66.94	67.21	66.87	66.84	66.59	67.48	67.41	67.10	67.27	67.45
OCFS	60.90	63.43	64.41	64.39	64.96	60.28	61.49	62.69	62.52	62.69
OCFSX	63.91	65.62	65.56	65.65	65.85	65.82	66.11	66.19	66.55	66.92
DFPFS	57.20	57.04	57.03	57.05	57.03	61.56	61.39	61.27	61.44	61.17
DFPFSX	62.55	63.71	63.67	63.40	63.40	64.92	65.42	65.37	65.46	65.53