Abstract—Permission structure of Android applications
introduces security vulnerabilities which can be readily exploited
by third-party applications. We address certain exploitability
aspects by means of neural networks, effective classification
techniques capable of verifying the application categories. We
devise a novel methodology to verify an application category by
machine-learning the application permissions and estimating
likelihoods of the extant categories. The performance of our
classifier is optimized through the joint minimization of false
positive and negative rates. Applying our modus operandi to
1,700 popular third-party Android applications and malwares, a
major portion of the category declarations were judged
truthfully. This manifests effectiveness of neural network
decision engines in validating Android application categories.