Machine Learning Based App-Collusion Detection in Smartphones

The App-collusion is a multi-app attacking model in which two or more apps work together to achieve a malicious goal. Single app analyses tools are not able to detect app-collusion because they focus on single apps. This paper presents a novel and simple technique to detect collusive app-pairs using machine learning classifiers. The method employs a two-stage classification model. In the first stage, we have logistic regression based classifier to detect single malicious applications. In the second stage, we use the parameters obtained from the first stage to detect collusive app-pairs. In the second stage, we use either logistic regression or perceptron model. Our two-stage classification model is successful in detecting collusive app-pairs.