Deep Learning Approaches for Predictive Masquerade Detection

Table 7

The results of our experiments.

Dataset	Data Configuration	Model	Evaluation Metrics ()
			Accuracy	Precision	Recall	F1-Score	Hit	Miss	FAR	Cost	BDR	BTNR	g-mean	MCC
SEA Dataset	SEA	DNN	98.08	76.26	84.85	80.33	84.85	15.15	1.28	22.83	76.25	99.26	91.52	79.45
		LSTM-RNN	98.52	82.30	86.58	84.39	86.58	13.42	0.90	18.83	82.33	99.34	92.63	83.64
		CNN	98.84	87.77	87.01	87.39	87.01	12.99	0.59	16.51	87.72	99.37	93	86.78
	SEA 1v49	DNN	96.54	99.98	96.43	98.17	96.43	3.57	0.48	6.47	99.98	52.04	97.96	70.64
		LSTM-RNN	97.86	99.98	97.79	98.87	97.79	2.21	0.38	4.48	99.98	63.70	98.7	78.74
		CNN	98.78	99.99	98.74	99.36	98.74	1.26	0.19	2.40	99.99	75.51	99.27	86.22
Greenberg Dataset	Greenberg Truncated	DNN	93.97	92.23	80.67	86.06	80.67	19.33	2.04	31.57	92.22	94.41	88.89	82.53
		LSTM-RNN	94.72	94.88	81.53	87.70	81.53	18.47	1.32	26.39	94.87	94.68	89.7	84.76
		CNN	95.43	96.16	83.53	89.40	83.53	16.47	1.0	22.47	96.16	95.24	90.94	86.86
	Greenberg Enriched	DNN	97.57	96.92	92.40	94.61	92.40	7.60	0.88	12.88	96.92	97.75	95.7	93.08
		LSTM-RNN	97.98	97.57	93.60	95.54	93.60	6.40	0.70	10.60	97.56	98.10	96.41	94.28
		CNN	98.60	98.55	95.33	96.92	95.33	4.67	0.42	7.19	98.55	98.61	97.43	96.03
PU Dataset	PU Truncated	DNN	81.0	99.59	78.61	87.86	78.61	21.39	2.25	34.89	99.59	39.49	87.66	54.63
		LSTM-RNN	82.19	99.69	79.89	88.70	79.89	20.11	1.75	30.61	99.68	41.10	88.6	56.46
		CNN	83.75	99.74	81.64	89.79	81.64	18.36	1.50	27.36	99.73	43.38	89.68	58.79
	PU Enriched	DNN	90.44	99.84	89.21	94.23	89.21	10.79	1.0	16.79	99.84	56.72	93.98	70.64
		LSTM-RNN	91.31	99.88	90.18	94.78	90.18	9.82	0.75	14.32	99.88	59.08	94.61	72.61
		CNN	93.75	99.92	92.93	96.30	92.93	7.07	0.50	10.07	99.92	66.78	96.16	78.52