A Proposed Architecture Based on CNN for Feature Selection and Classification of Android Malwares

Malware detection process is based principally on extracting data given to classifier model; those data are information about application’s behavior during its execution, permissions required by it or activities made in response to some commands. Which clearly make the features chosen and build as features vector highly influence the credibility of the model in classifying with high accuracy the unknown applications. For this reason, the research field gave a decent attention to resolve this problematic in malware detection models by improving the quality of features used in classification process, and performing feature selection processes in order to reduce dimensionality of features vectors, selecting most relevant, correlated and informative features and to eliminate redundant information. Many solutions were invented for this purpose using machine-learning algorithm to evaluate performance of classification using a specific set of features or by using filter feature selection algorithms that give a rank to each feature depending on its occurrence frequency, weight or its correlation. In this paper, we proposed an approach using CNN deep learning model for classifying and detecting android malwares as a solution for feature selection and redundancy problematic.