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

Table 6

The comparison of nine improved and existing feature-selection methods with respect to AUC 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	0.8978	0.9068	0.9073	0.9088	0.9093	0.9005	0.9050	0.9065	0.9079	0.9086
IGX	0.8977	0.9068	0.9073	0.9087	0.9093	0.9083	0.9124	0.9143	0.9159	0.9168
CHI	0.8988	0.9055	0.9093	0.9091	0.9112	0.9053	0.9071	0.9066	0.9070	0.9079
CHIX	0.8864	0.9058	0.9074	0.9101	0.9116	0.9090	0.9126	0.9143	0.9144	0.9138
MI	0.6923	0.8521	0.8799	0.8902	0.8979	0.7957	0.8541	0.8810	0.8904	0.8974
MIX	0.7282	0.8256	0.8815	0.8928	0.9007	0.7357	0.8341	0.8897	0.8970	0.9070
DF	0.8977	0.9075	0.9098	0.9107	0.9107	0.9012	0.9060	0.9091	0.9086	0.9090
DFX	0.9008	0.9091	0.9105	0.9111	0.9116	0.9103	0.9138	0.9159	0.9169	0.9167
GINI	0.9082	0.9119	0.9135	0.9138	0.9123	0.9083	0.9068	0.9075	0.9088	0.9083
GINIX	0.9112	0.9129	0.9139	0.9135	0.9135	0.9165	0.9147	0.9152	0.9157	0.9164
DIA	0.7005	0.7146	0.7249	0.7334	0.7565	0.8501	0.8673	0.8750	0.8790	0.8811
DIAX	0.7884	0.8654	0.9040	0.9066	0.9088	0.7954	0.8778	0.9124	0.9143	0.9159
CMFS	0.9109	0.9141	0.9133	0.9139	0.9134	0.9095	0.9094	0.9084	0.9087	0.9094
CMFSX	0.9071	0.9135	0.9143	0.9147	0.9144	0.9150	0.9171	0.9153	0.9156	0.9162
OCFS	0.8983	0.9074	0.9088	0.9092	0.9104	0.9027	0.9029	0.9046	0.9057	0.9075
OCFSX	0.8914	0.9065	0.9091	0.9095	0.9102	0.9068	0.9094	0.9111	0.9127	0.9136
DFPFS	0.8159	0.8157	0.8159	0.8161	0.8162	0.8875	0.8869	0.8865	0.8871	0.8863
DFPFSX	0.8828	0.8882	0.8884	0.8880	0.8883	0.9032	0.9066	0.9064	0.9062	0.9067