Malware Detection on Byte Streams of PDF Files Using Convolutional Neural Networks

Table 4

Experimental results with different settings, where the two values of each cell are of ‘benign’ and ‘malicious’, respectively.

Dimensions & Training options	Precision	Recall	F1
E = 15	99.68 / 100.0	92.52 / 95.75	95.34 / 97.82
E = 35	99.73 / 100.0	94.03 / 97.47	96.58 / 98.71
Drop-out with 0.5	99.70 / 100.0	93.53 / 97.37	96.23 / 98.66
L2 gradient-clipping with 0.3	99.74 / 100.0	95.25 / 97.17	97.17 / 98.55
L2 gradient-clipping with 0.7	99.70 / 100.0	92.11 / 97.78	95.14 / 98.86
w/o batch-normalization	99.70 / 100.0	95.96 / 97.27	97.68 / 98.61
P = 50	99.70 / 100.0	95.05 / 97.07	97.13 / 98.50
P = 150	99.76 / 100.0	93.83 / 97.88	96.29 / 98.92