HRS

Abstract

Traditional signature-based detection methods fail to detect unknown malwares, while data mining methods for detection are proved useful to new malwares but suffer for high false positive rate. In this paper, we provide a novel hybrid framework called HRS based on the analysis for 50 millions of malware samples across 20,000 malware classes from our antivirus platform. The distribution of the samples are elaborated and a hybrid framework HRS is proposed, which consists of Hash-based, Rule-based and SVM-based models trained from different classes of malwares according to the distribution. Rule-based model is the core component of the hybrid framework. It is convenient to control false positives by adjusting the factor of a boolean expression in rule-based method, while it still has the ability to detect the unknown malwares. The SVM-based method is enhanced by examining the critical sections of the malwares, which can significantly shorten the scanning and training time. Rigorous experiments have been performed to evaluate the HRS approach based on the massive dataset and the results demonstrate that HRS achieves a true positive rate of 99.84% with an error rate of 0.17%. The HRS method has already been deployed into our security platform.

Malware Detection and Analysis Tools

“…There have been many studies on the detection and analysis of malware using machine learning that study fine-grained features [34], deep learning [35][36][37], dynamic features [38], static fea-tures [36,39], concept drift [40], predicting signatures [41], hybrid framework [42], malware metadata [43], reverse engineering of large datasets of binaries [44].…”