Mobile malware detection using deep neural network

Malware has become more harmful than in the past as the number of intelligent systems and Internet-connected devices increased dramatically. Therefore one of the most important issues in cyber security has become the detection of previously unknown malware in the shortest time possible in order to stop it from becoming epidemic and from harming users. Antivirus (antimalware) systems using signature databases of previously identified malware quite successfully identify existing malware but they are far from achieving the same detection performance for new malware. Consequently, machine learning methods were applied to determine and classify malware. For machine learning algorithms to achieve a better performance it is necessary to run the malware in a sandbox to collect features from the malware which can not be obtained statically. However, this fact gave the malware authors the upper hand in that they developed special anti-analysis techniques to mislead the machine learning based analysis in that they detected the sandbox environment and changed their malicious behavior to a normal behavior. Therefore in this study, we present a novel model based on deep learning for the prediction of mobile malware without requiring execution in a sandbox environment. Application permissions were used as features. After optimizing their weights with automatic encoder and they were classified with a multilayer perceptron with an accuracy of 93.67%.