Android Malware: Comprehensive Study and a Cross-Feature Light Weight Proposed Solution

The need for adequate safety precautions for Android-based gadgets has grown due to the fast expansion of applications for the platform and the sophisticated nature of threats such as malware. The ever-changing world of Android malware has been challenging to identify and categorize using conventional signature-based methods. To strengthen Android security, experts have resorted to advanced methodologies such as federated learning, machine learning, ensemble learning, deep learning, and natural language processing techniques. This paper studies the literature of the decade 2013-2023. After thoroughly investigating existing literature, this paper presents the most recent advancements and emerging trends in the domains of Android security. This study also presents a comprehensive cross-analysis of various datasets that researchers use. Also, suggestions for building new datasets have been proposed in this paper. This study will direct future researchers to build cutting-edge mechanisms for the detection of unknown malware using lightweight classifiers which saves computation time and resources. This paper also proposes a novel approach for identifying malicious applications and it makes use of the Bi-directional Transformer-based model (BERT) for feature extraction and employing a lightweight classifier for efficient classification, optimizing accuracy, and reducing computing costs.