A Comprehensive Review of Feature Selection Strategies for Machine Learning Based Android Malware Detection

The emergence of even more advanced attacks of mobile malware that have been trying to steal user credentials and other important data has also been accompanied by the concept of mobile smartphone use that is projected to be in the billions across the globe. With high dimensionality of features in modern malware datasets, effective feature selection is important to perfor-mance as well as efficiency of machine learning (ML) models, which often involve selecting most relevant features from a potentially large set of features. This process not individual reduce computational overhead but also increases the accuracy and generalization of models, which can overcome the challenges associated with more complex and numerous malicious software instances. Feature selection is an important step in designing accurate and efficient Android malware detection systems because of high dimensionality of data obtained from static and dynamic application analysis. This problem is compounded by the fact that cyber-criminals are constantly inventing new ways to obscure their attacks, and signature-based detection is no longer sufficient, which has led to need for more dynamic ML approaches.