Evaluating Convolutional Neural Network for Effective Mobile Malware Detection

摘要: In last years smartphone and tablet devices have been handling an increasing variety of sensitive resources. As a matter fact, these store plethora information related to our every-day life, from the contact list, received email, also position during day (using not only GPS chipset that can be disabled but Wi-Fi/mobile connection it is possible discover device geolocalization).This reason why mobile attackers are producing large number malicious applications targeting Android (that most diffused operating system), often by modifying existing applications, which results in malware being organized families, where each application belonging same family exhibit behaviour. These behaviours typically gathering related, for instance very widespread behaviour represented sending personal (as examples: send SMSs, browser history) remote server managed attackers.In this paper, we investigate whether deep learning algorithms able discriminate between legitimate samples. To end, designed method based on convolutional neural network applied syscalls occurrences through dynamic analysis. We experimentally evaluated built classifiers recent dataset composed 7100 real-world more than 3000 several different families order test effectiveness proposed method, obtaining encouraging results.