Dro-Mal Detector: A Novel Method of Android Malware Detection

The increasing popularity of Android applications accompanies with it an ever-growing field of malware. Hence, the need of the hour is to design efficient malware detection techniques to ensure security. The API call graph of an Android Application Package (APK) is an efficient way to understand the behavior of an application and caller-callee relationships. However, traditional algorithms of graph similarity are not optimized enough to perform classification for large Android Application packages (APKs). Our study aims to use the function call graph of an API as a model to track malware behavior statically and dynamically. The call graph embeddings and permissions when converted to reduced dimensional feature vectors, are then trained with the help of Support Vector Classifier (SVC) whose hyperparameters are tuned with the help of the Grid Search algorithm. The same dataset is also trained with Artificial Neural Networks (ANNs) and compared by varying different hyperparameters. The results of both algorithms were compared and it was found that the Support Vector Classifier (SVC) performed better than the deep learning algorithm.