A Trustworthy Convolutional Neural Network-Based Malware Variant Detector in Python

Abstract

Malware attacks are broadly disguised as useful applications. Many android apps, downloaded to perform crucial tasks or play games (take one's pick), seem to do completely different tasks, which are potentially harmful and invasive in nature. This could include sending text messages to random users, exporting the phone's contacts, etc. There exist some algorithms in place that can detect these malwares, but so far, it has been observed that many of these algorithms suffer from false negatives, which grossly reduced the effectiveness of said algorithms. The aim of this chapter is to introduce a flexible method to detect if a certain application is malware or not. The working can be loosely defined as the source of a set of applications is detected and the list of permissions is studied. The set of relevant and highly close applications is selected, and from the most relevant category, the permissions are checked for overlap to see if it can be stated as a possible anomalous application.