Minimization of Attributes for the Detection of Vulnerabilities in Android Applications

Abstract

Bluetooth, MMS, SMS, e-mail, and other mobile-specific applications may constitute deliberate threats, resulting in financial losses as well as the revealing of sensitive user data. To circumvent this, different approaches have lately been applied, such as static analysis-based detection or dynamic analysis-based detection. However, some of these strategies have become saturated as a result of massive increase in malware code authoring techniques. As a result, there is a need for an effective android privacy-leakage vulnerability detection technique, for which this research used hybrid features such as standard permissions, non-standard permissions, system call traces, and network traffic to conduct experimental tests. We presented a method to prioritize the attributes in order to select the most prominent features, with the purpose of minimizing the number of attributes to be investigated while retaining good detection accuracy. To rank the attributes in this dimension, we used a variety of statistical approaches. The findings of the experiments reveal that selecting attributes for vulnerability detection resulted in higher detection accuracy than assessing Accuracy using all attributes. To assess the efficacy of the suggested attribute selection procedure, we deployed a neural network. With the prominent attributes chosen by the proposed algorithm, the neural network attained an accuracy of 96.41 percent, which is 2% higher than the accuracy we achieved with all of the attributes combined.