A novel end-to-end ensemble framework for enhanced android malware detection accuracy

The accelerated growth in the use of the Android platform globally has positioned it as a prime target for cybercriminals. The malware creators meticulously and prudently craft fraudulent code with a range of capabilities, including spying, stealing credentials, and unauthorised system access. As a result, detecting harmful behaviour is highly challenging, stressing the importance of studying malware detection. Several proposals have recently surfaced for malware detection through the use of machine learning algorithms. However, many ideas lack the extraction of representative features, resulting in intrinsic flaws such as inaccuracy, low efficiency, and insufficient robustness to detect current malware. In this article, we propose E3Droid, an end-to-end ensemble framework to detect Android malware. The study involves extracting features from ensemble analysis, including static, visual, and dynamic analysis. An ensemble feature selection method is applied to choose the optimal features, combining Relief, Chi-square, and MRMR algorithms. In addition, an ensemble classifier consisting of basic learners SVM, KNN, and NB algorithms, is employed. The proposed E3Droid framework is thoroughly evaluated and compared with state-of-the-art techniques through extensive experimentation. It showcases outstanding results in all assessment criteria, attaining an accuracy level of 99.80% and a false positive rate of 00.24%.