Malware Detection with Structural Entropy Features Using Multilayer Perceptron Neural Network

With the increasing number of computer devices, malware poses a threat to the current information security. Malware writers employ various obfuscation techniques to evade traditional signature-based detection. This paper proposes a pre-liminary study of malware detection using Multilayer Perceptron Neural Network with structural entropy feature. A deterministic algorithm is used to generate these features that represent the feature signature of an executable file. Additionally, we investigate the effectiveness of different byte sizes as the input feature for the neural network. Our preliminary results show there exist performance variability between the 3-layer model and 4-layer model on different source of malware datasets.