Phishing Target Identification Based on Neural Networks Using Category Features and Images

Table 3

The performance of different features’ sets on LightGBM.

Feature set	Accuracy	Macro-F1 score	Weighted-F1 score
+  +	0.8325	0.6655	0.8197
	0.8743	0.7005	0.8659
	0.8778	0.7115	0.8692
	0.8813	0.7313	0.8768
	0.7958	0.6375	0.7771
	0.8255	0.6647	0.8103
	0.8674	0.6829	0.8581
	0.7941	0.6576	0.7774
	0.8743	0.7115	0.8680
	0.8551	0.7415	0.8521
	0.7644	0.5805	0.7418
	0.7365	0.5838	0.7104
	0.6143	0.5067	0.5849
	0.2583	0.0064	0.1172

represents URL features, represents host features, represents web resource features, and represents OCR features. The bold values represent the best values of the evaluation metrics.