A Comparative Analysis of Android Malware Detection Using Deep Learning

The fear of Android malware infection is building up with its user acceptance in the global market space. The existing anti-malware techniques are archaic to signature-based detection which limits the detection scope against newly crafted malware. Several machine learning, as well as deep learning approaches, has been proposed so far to combat against the rising threat of the android mobile world. Most of these have relied on static features due to its lower cost. In this paper, a Deep Neural Network model is proposed and evaluated it with static, dynamic and hybrid features respectively to prepare the comparative statement. In this experiment we used the OmniDroid dataset [1] which consists of 22,000 Android Package Kits (APKs) including both benign and malware APKs. For analysis 7955 static features and 4805 dynamic features are extracted using proper selection criteria. The efficacy of the proposed models on different data types is compared in terms of accuracy and loss. This experiment achieved a higher accuracy of 99.66% with dynamic features and a lower accuracy of 85.74% with combined features. Similarly, we achieved a minimal loss of 0.06 with dynamic analysis and a higher loss of 0.61 with static analysis. This confirms that the efficacy of dynamic analysis is more prominent in An- droid malware detection.