Deep Learning na segurança computacional: detecção inteligente de códigos maliciosos

Abstract

The increase in the amount of malware and their families amplified the problems of automatic detection and classification of their new variants. As computer security threats evolve, so does the need for effective defense mechanisms to protect the devices. However, it becomes progressively more difficult to protect terminals from being infected. Thus, tools which identify resident malicious codes are required for handling post-infection systems. In this work of course completion, the application of a Deep Neural Network (DNN) architecture to detect malwares based on its operational system processes is investigated. The Deep Learning framework proposed implements a AutoEncoder and utilizes API call sequences to extract features, forming vectors that function as signatures of malicious codes. Samples of malicious and benign codes were obtained to train and test the classifiers. The effectiveness of AutoEncoder built to facilitate the correct classification of the malicious codes was made evident by the results obtained from the classifiers.