Machine Learning Approaches for Analysing Static features in Android Malware Detection

Abstract

As the number of people who own and use smart-phones grows, attackers are always looking for ways to steal sensitive information from mobile phones. Researchers are always working on making it easier to find android malware so that information can be hidden and kept safe. Since the number of new malware is going up, techniques based on machine learning are the best way to find them on a large scale.CICInvesAndMal2019 included in this paper uses android permissions and intent as a dataset and a set of features to look for malware. As a way to choose features, Principal Component Analysis (PCA) was used. Different machine learning (ML) models like Naive bias, Decision tree (DT), and Random Forest(RF), k-NN are used to train and test the dataset. The dataset is modeled and evaluated on well-known ML models, and RF was the best classifier in binary classification with a 99.7% success rate and in case of category classification RF was the best classifier with 97.30% success rate for ransomware category.