With the sudden upswing in digitization, usage of Android smartphones has evidently become conventional and common. Nevertheless, such extensive popularity of Android systems comes with a major trade-off between usage and security. Time and again, it has been proven that the existence of malicious applications leverage permissions given by the Android systems in order to compromise them. However, there exist certain combinations of permission sets that could potentially lead to such attacks. To address this, we propose a two-layer architecture, where each layer has an implicit handshake with a Machine Learning ensemble model, that aims at vulnerability detection as well as classification of these detected vulnerabilities into one among the three levels of severity – High Risk, Medium Risk and Low Risk. Efficiency of the proposed system has been evaluated against existing Machine Learning based models such as K-Nearest Neighbors (KNN), Decision Tree Algorithms and Multi-Layered Perceptron (MLP) on an Android Permission-Based Dataset. Results show that our proposed model outperforms the existing approaches, which has been quantitatively evaluated using accuracy and F1-score.

No abstract