Estudo e implementação de classificadores binários para detecção de malwares de Android baseados em features estáticas

This work provides a comprehensive overview of the growing dependence of contemporary society on technology, particularly mobile devices. In light of this, it is worth noting that malicious Android applications can be used for a wide range of illicit activities that impact the security and privacy of users. Therefore, intervention by researchers is necessary to address and mitigate this type of cyber threat. To contribute to this premise, a systematic literature review was conducted, covering studies from 2015 to 2021, to identify techniques and tools that address malware analysis issues on Android devices. As a result, 60 articles were read in full to compile and categorize the most recurring techniques and tools, malware types, and anti-malware analysis techniques. Consequently, the results of this work lead to the conclusion that malware analysis on Android devices is continually evolving, with techniques available both for analyzing samples and for preventing such analysis. Furthermore, 118 techniques from different classes were identified to address the problems in this area of study, along with 357 tools categorized according to the techniques, 9 anti-analysis techniques, and 14 types of malware. Additionally, by assembling a dataset of 10,000 applications, both legitimate and malicious, and using Permission-based Analysis, Call Graph Analysis, and Taint Analysis techniques, it is possible to characterize Android applications by identifying reachable methods in an app's call graph and methods that leak sensitive user information. With the dataset ready, the following binary classifiers were implemented and evaluated: Decision Tree, Random Forest, Adaboost, Naive Bayes, and SVM (RBF).