phishGILLNET—phishing detection methodology using probabilistic latent semantic analysis, AdaBoost, and co-training

EURASIP Journal on Information Security

Table 6 phishGILLNET2--3-Class (phish versus spam versus good) classification performance

Topics	Weak learner for boosting	TPR	FPR	Precision	Recall	F-measure	ROC Area	Time (s)
50	C4.5	0.954	0.088	0.954	0.954	0.954	0.944	1.84
50	RIPPER	0.964	0.069	0.964	0.964	0.964	0.955	12.07
50	Random forest	0.974	0.079	0.973	0.974	0.973	0.996	3.09
50	SVM	0.91	0.199	0.907	0.91	0.908	0.867	12.41
50	Logistic	0.909	0.238	0.905	0.909	0.905	0.957	2.42
100	C4.5	0.967	0.068	0.967	0.967	0.967	0.961	5.06
100	RIPPER	0.974	0.043	0.975	0.974	0.975	0.971	16.6
100	Random forest	0.976	0.075	0.975	0.976	0.975	0.997	3.31
100	SVM	0.964	0.095	0.964	0.964	0.963	0.94	11.32
100	Logistic	0.971	0.065	0.97	0.971	0.97	0.989	5.05
200	C4.5	0.969	0.061	0.969	0.969	0.969	0.961	8.93
200	RIPPER	0.972	0.048	0.973	0.972	0.972	0.968	24.77
200	Random forest	0.977	0.06	0.977	0.977	0.977	0.996	3.7
200	SVM	0.97	0.071	0.971	0.97	0.97	0.953	18.62
200	Logistic	0.971	0.065	0.97	0.97	0.97	0.989	6.15