A visualization-based analysis on classifying android malware

Since the introduction of the Android mobile platform, the state of mobile malware has evolved in both attack sophistication and its ability to evade detection. Given the right combination of elements, the detection of malicious applications may be found among those that pose no threat, yet the threats that exist across these malware types reveal distinguishable attack characteristics. This paper investigates the benign and attacking characteristics. By plotting complex features into dendrograms, we propose a novel approach to visually distinguish Android apps. We visualize the complicated relationship and evaluate the effect of different text mining methods. Specifically, we employ machine learning techniques including feature reduction using Principle Component Analysis, and the Random Forest classifier, to compare eight different models. Using the Drebin dataset, we achieved an average accuracy of 95.83%.