A Deep Learning Method for Android Application Classification Using Semantic Features

Table 6

Results obtained using different feature sets.

Region size	Train time (s)	ACC (%)	PRE (%)	REC (%)	F1 (%)
Feature set I
(3)	37	99.0905	98.9735	99.1794	99.0763
(5)	39	99.0508	98.3917	99.6995	99.0413
(3,4,5)	63	99.1190	99.0872	99.1216	99.1044
(2,3,4,5)	80	99.0962	98.7040	99.4683	99.0847
(3,3,3)	62	99.0280	99.2337	98.7864	99.0096
(3,3,3,3)	74	99.1303	98.7719	99.4683	99.1189
(7,7,7)	81	99.0621	98.9616	99.1331	99.0473
Feature set I + II
(3)	36	99.1995	99.0590	99.3307	99.1947
(5)	38	99.1886	98.7705	99.6050	99.1860
(3,4,5)	63	99.2213	99.2966	99.1332	99.2148
(2,3,4,5)	76	99.2430	99.1781	99.2978	99.2379
(3,3,3)	58	99.1886	99.2096	99.1551	99.1824
(3,3,3,3)	71	99.2267	98.9524	99.4953	99.2231
Feature set I + II + III
(3)	43	99.3100	98.8877	99.7360	99.3100
(5)	44	99.2771	99.0153	99.5381	99.2760
(3,4,5)	70	99.3538	99.0061	99.7030	99.3534
(2,3,4,5)	85	99.2552	99.0794	99.4281	99.2534
(3,3,3)	69	99.2662	99.0044	99.5271	99.2650
(3,3,3,3)	79	99.3045	98.9089	99.7030	99.3044