A deep learning based model for detection of Android malwares using PCA over physical devices
Bilal Ahmad Mantoo · AIP conference proceedings · 2022
The Android operating system is considered to be the most advanced and popular smartphone operating system and has a striking enhance in the Android Platform with approximately of 1.5 billion Android-based devices shipped by 2021.This spike increase in the market leads to a high number of malware in the Android platform. The reason is simple because of the mutating nature of Android malware reduces the efficiency of malware detection tools in the market. Furthermore, the vast number of features in the Android platform provides a big challenge to the current tools to find Android malware. In this paper, a real physical device i.e. Android smartphone is used instead of a protected environment like gennymotion for analysis and extract the feature from 10650 applications of malware and benign. Large spaces of feature set are reduced with the use of the Principal Component Analysis approach which extracts the best features from the dataset. The data is then fed to the Deep Learning model with different hidden layers; the algorithms are applied for dynamic features as well as the combination with permissions. The results reveal that Deep learning model with three hidden layers gives the overall best accuracy as compared to the previous works done on the same field using a protected environment.