Abstract:
Nowadays, we see a good number of mobile phone users using Android applications, and hence
we witness more and more android applications being developed. As having the apps makes life
easier, it also opens the door for many vulnerabilities as well. According to an estimate, Android
covers around 85% of the smartphone volume resulting in more than 95% of the mobile malware
targeting Android applications and devices [1] [2] [3]. Therefore, there is a requirement for a
sophisticated malware detection system that could detect or predict malware in the android
applications in android smartphones.
We find significant work done in this field using state-of-the-art technologies and methods for
efficient detection or prediction of malware. However, malware developers are also becoming
smart enough to bypass those solutions and penetrate the Android applications. Some researchers
have proposed the hybrid models, and some proposed the cloud-based models to detect and predict
the malware applications efficiently, but there are significant latency issues. We can overcome this
delayed addition of new signatures and examples by using Generative Adversarial Networks
(GAN) that helps train a generative part by treating malware detection as a supervised machine
learning problem having two sub-models, i.e., generator and discriminator model. The generator
model gets used to generate a new model, whereas the discriminator gets used to classify examples
from domain or generated ones, also called fake examples.
We consider four types of malware i.e. Adware, Ransomware, Backdoors, and Trojans, and we
produce counter-examples to increase the Model's efficiency. Our Model has two parts,
Generative and discriminative. The generative part learns on the dataset, then generates the
applications based on its learning. Then the discriminator part predicts if the generated example
is malware or benign based on its training on the dataset given to it previously.
