ShielDroid: A hybrid approach integrating machine and deep learning for android malware detection

Abstract

Due to the rapid development of the advanced world of technology, there is a high increase in devices such as smart-phones and tablets, which increase the number of applications used. Though an application has to pass the malware detection test before appearing in the play store, many applications successfully get trusted and accepted even though they contain malicious software variants that are challenging to detect. The application requires physical execution to see these malicious contents, which get undetected during the first screening test. Due to the physical implementation of the application, it may be too late to undo the malware's damage. In this work, the usage of real-time Android malware detection analyzing Android applications to detect and swiftly distinguish complex malware has been discussed. This work focuses on the use of dynamic algorithms implemented by hybrid detection techniques of Android malware. After filtrating the collected dataset, the process of separation between harmful and benign apps is discussed. Then summarization and evaluation of the various techniques and classification algorithms employed have been discussed, identifying the best-suited method that gives the most accurate result in a minimum amount of time. The best way to reach the target is a hybrid Random Forest, and Multilayer perceptron network, where the overall accuracy achieved was 97.5% with an execution time of 22.945 seconds. The application of this work may allow the users to protect their devices from cyber-attacks by detecting malicious software variants in mobile applications.