Malicious software behavior detection and classification system based on deep learning

The invention discloses a malicious software behavior detection and classification system based on deep learning. The malicious software dynamic behavior data captured based on the sandbox comprises two core components of API calling sequence feature engineering and model construction. The feature engineering part gets rid of the defects that in previous research, only attribute features are focused, structural features are neglected, attribute features are collected mainly based on N-gram, and then semantic understanding ability is lacked. The model construction part is used for constructing a comprehensive detection and classification model by utilizing an integrated learning thought; the sub-models are selected and designed based on supervised learning, unsupervised learning and reinforcement learning so as to realize high-precision detection of conventional malicious software and high-precision detection of unknown APT malicious software; an adversarial training method is adopted, adversarial samples gene