Enhancing Android Security: Dynamic Analysis for Resilient Defences Using Machine Learning

In recent years, the widespread popularity of Android as a mobile operating system has been accompanied by an abundance of mobile applications, enriching the user experience. However, this surge in Android app availability has also ushered in a heightened risk of encountering malicious Android applications, commonly known as malware, particularly when downloading from third-party sources. The formidable challenge lies in the efficient and precise detection of such malware. In this research, a unique virus detection approach optimized for the Android platform is studied, harnessing the potential of dynamic analysis. Through rigorous experimentation, the performances are evaluated of the proposed scheme, showcases its effectiveness and efficiency in identifying malicious Android applications. Notably, a diverse set of machine learning models are employed and their respective accuracies are calculated, these models include K-Nearest Neighbors (KNN) with 0.8825%, Naive Bayes with 0.4409%, Gradient Boosting with 0.8524%, Random Forest with 0.9122%, XGBoost with 0.8907%, LightGBM with 89.07% and Bagging 0.9040%, with Random Forest giving the best accuracy and Naive Bayes the least. the study not only delves into the intricacies of dynamic analysis but also highlights the accuracy scores achieved by these prominent machine learning models, providing valuable insights for the field of Android malware detection.