Analyzing Various Machine Learning Approaches for Detecting Android Malware

The Android operating system is the world's largest smartphone platform. The platform carries with it significant risk due to the vast amounts of malware that target the platform. Newer methods involve the use of machine learning to detect malicious software in a more general manner, as opposed to conventional methods that rely on prior knowledge pertaining to a particular malware family. This study explores the use of a Convolutional Neural Network and various tree-based learning methods processing feature engineered from malware binaries. Overall model accuracy was found to be between 87% and 90%, and sufficient performance was obtained when collecting a certain amount of data above the threshold. It was also found that tuning some of the tree based model hyperparameters caused overfitting. Overall, the findings concluded that models that did not overfit performed comparably during both training and testing, representing consistent detection models.