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

Table 4

The comparison of nine improved and existing feature-selection methods with respect to AUC on 20-Newsgroups 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	0.7183	0.7579	0.7864	0.8048	0.8187	0.7746	0.7998	0.8144	0.8257	0.8318
IGX	0.7182	0.7578	0.7863	0.8045	0.8187	0.7774	0.8038	0.8183	0.8284	0.8361
CHI	0.8234	0.8545	0.8660	0.8717	0.8771	0.8485	0.8671	0.8702	0.8707	0.8690
CHIX	0.8346	0.8621	0.8710	0.8770	0.8815	0.8589	0.8726	0.8747	0.8763	0.8753
MI	0.5830	0.6283	0.7205	0.7405	0.7695	0.6165	0.6733	0.7569	0.7721	0.7905
MIX	0.6020	0.6597	0.7315	0.7677	0.7902	0.6269	0.6894	0.7700	0.7961	0.8089
DF	0.7685	0.8054	0.8302	0.8454	0.8531	0.8043	0.8224	0.8380	0.8464	0.8509
DFX	0.7763	0.8111	0.8338	0.8501	0.8585	0.8146	0.8297	0.8416	0.8500	0.8557
GINI	0.8569	0.8726	0.8778	0.8801	0.8829	0.8711	0.8678	0.8631	0.8609	0.8625
GINIX	0.8603	0.8749	0.8800	0.8843	0.8858	0.8720	0.8735	0.8690	0.8660	0.8672
DIA	0.5571	0.6040	0.6107	0.6165	0.6245	0.6331	0.6838	0.7086	0.7302	0.7419
DIAX	0.7429	0.7933	0.8219	0.8577	0.8728	0.7519	0.8038	0.8324	0.8585	0.8690
CMFS	0.8536	0.8671	0.8748	0.8828	0.8865	0.8689	0.8769	0.8747	0.8723	0.8705
CMFSX	0.8598	0.8736	0.8821	0.8887	0.8915	0.8722	0.8798	0.8788	0.8776	0.8766
OCFS	0.7083	0.7375	0.7892	0.8017	0.8113	0.7962	0.8164	0.8260	0.8301	0.8351
OCFSX	0.7550	0.7882	0.8091	0.8205	0.8290	0.7984	0.8193	0.8313	0.8365	0.8408
DFPFS	0.8175	0.8398	0.8499	0.8561	0.8600	0.8377	0.8446	0.8492	0.8521	0.8543
DFPFSX	0.7591	0.7873	0.8000	0.8056	0.8086	0.7952	0.8119	0.8181	0.8218	0.8240