Intelligent Behavioral Monitoring and Detection of Preinstalled Malicious Applications on Smartphones Sold in Emerging African Markets

This paper addresses the problem of detection preinstalled malicious mobile applications on low-cost smartphones sold to emerging african markets. Using primary data consisting of 1427 mobile applications extracted from 6 Android smartphones, a static approach is employed to extract and analyze application technical features, including permissions, certificates, network traffic and manifest. Learning models based on ensemble learning have been designed and tested using Bagging, Boosting and Stacking methods. Assessment results shows that the Stacking method, combining the SGD, KNN and DT classifiers, provides the best performance. Indeed, Stacking method delivers a detection rate of 87.1 %.