Malware Classification Using Euclidean Distance and Artificial Neural Networks
Lilia E. Gonzalez, Roberto Antonio Vazquez · 2013
Most of the samples discovered are variations of known malicious programs and thus have similar structures, however, there is no method of malware classification that is completely effective. To address this issue, the approach proposed in this paper represents a malware in terms of a vector, in which each feature consists of the amount of APIs called from a Dynamic Link Library (DLL). To determine if this approach is useful to classify malware variants into the correct families, we employ Euclidean Distance and a Multilayer Perceptron with several learning algorithms. The experimental results are analyzed to determine which method works best with the approach. The experiments were conducted with a database that contains real samples of worms and trojans and show that is possible to classify malware variants using the number of functions imported per library. However, the accuracy varies depending on the method used for the classification.