Malware Detection Based on Deep Learning of Behavior Graphs

Table 3

The comparison between our study and other deep learning methods.

Method	Features	Machine Learning Model	Precision	Recall	F1-Score
William et al. [12]	API calls	Stacked AutoEncoders	0.955	0.958	0.956
Zhenlong et al. [44]	Permissions, Sensitive APIs, App actions	Deep Belief Networks	0.968	0.968	0.968
Toshiki et al. [45]	Malware communication	Recursive Neural Network	0.976	0.962	0.969
Proposed SAE-DT	Behavior graphs	Stacked AutoEncoders	0.986	0.992	0.989