PDRCNN: Precise Phishing Detection with Recurrent Convolutional Neural Networks

Table 9

Performance comparison with 4 baseline.

Method	Accuracy (%)	Precision (%)	Recall (%)	F-measure (%)	AUC (%)	Training time (s)	Test time (s)
PDRCNN	95.6	97.33	93.78	95.52	98.96	4426.15	40.66
RNN	94.24	95.14	93.26	94.19	98.33	2033.92	17.85
CNN	94.46	95.52	93.31	94.4	98.48	442.79	4.47
CANTINA+	76.43	78.24	76.98	76.44	85.43	133h	30h
9 features with GaussianNB	64.91	94.24	31.81	47.57	77.93	108.5	20.64
9 features with LG	71.2	75.37	63.05	68.66	78.21	107.1	20.81
9 features with GBDT	71.51	75.38	63.94	69.19	78.51	148.21	21.13
BiGram with GaussianNB	68.69	92.3	40.82	56.61	69.71	2985.94	638.52
BiGram with LG	88.68	90.96	85.91	88.36	95.68	2982.34	652.74
BiGram with GBDT	84.55	88.32	79.65	83.76	92.55	26420.99	639.12

Then, we use the ROC curve and the AUC value to evaluate the PDRCNN model and the four baseline models.