Development of an approach to the detection of Android software using deep learning methods

Abstract

The purpose of the work is to develop an approach to detect malicious software for the Android operating system based on statistical analysis using deep learning methods. To achieve the goal, the following tasks were solved: 1. Study of the features of Android applications and development of a method of submitting the application for further security analysis. 2. Research of deep learning methods and selection of the most appropriate of them. 3. Development of an Android malware detection approach using deep learning techniques. The main idea of the approach is to represent the Android application in the form of an image for further analysis by a convolutional neural network, and in this image the pixels represent a sequence of API call pairs and the level of protection against it, which is derived from the permission required for the API call. An Android malware detection approach is developed based on the representation of Android applications, as well as a convolutional neural network that has been specially developed for image recognition. A sequence of pairs of API calls and security levels of Android applications is converted into an RGB image, which is then fed to the input of a convolutional neural network. Having trained on a sample of similar images, the neural network acts as a classifier of included Android applications into legitimate and malicious ones.