A Deep-Vision-Based Multi-class Classification System of Android Malware Apps

The number of malicious software attacks on the Android operating system (OS) is increasing daily. Thus, efficient detection and classification models must be used to differentiate between benign and malware Android apps. Unfortunately, conventional malware detection and classification techniques based on traditional static- or dynamic-based machine learning (ML) algorithms are not the best choices for malware analysis applications. These traditional detection techniques are based on obtaining signature or behavior features using static or dynamic techniques. Therefore, using more intelligent and automated malware detection algorithms based on vision-based deep learning (DL) techniques for malware analysis is advised. Consequently, this chapter introduces a deep-vision-based multi-class classification system of Android malware applications. This proposed classification system composes 21 different DL algorithms for malware detection and recognition. The vision-based classification system was evaluated comprehensively using two open-source Android datasets (CICAndMal2017 and CICMalDroid2020). The binary formats of the android apps included in these datasets were first converted into color and grayscale vision formats before forwarding them to DL algorithms for training and testing mechanisms. In addition, the classification performance of the proposed vision-based detection system was examined using different security and recognition metrics. The obtained classification outcomes prove the high detection capability of the suggested multi-classification system in powerfully detecting various malware families in Android cybersecurity applications.