Leveraging Convolutional Neural Networks for Malware Detection: A Review of Techniques, Architectures, and Emerging Trends

Abstract

With the modernization of the Internet and society, people often use some mobile phone applications to facilitate their lives. But since most of these apps require sensitive information from users, malware has been developed by malicious people to infect users' phones, especially on the Android platform. In order to detect whether a file after download is malware, this paper takes a static detection method. By decompiling and uncompressing the downloaded APK file, extracting the software rights from it, converting the rights into feature graphs, and putting them in a neural network to test. After four fold cross-testing, the average accuracy on the validation set was 88.205 percent. The group that trained the network with the highest accuracy of its data sets was then selected as the training set for the final network, and then tested on a test set, which showed that this method worked better than traditional machine learning methods.