Abstract
Owing to the increasing number and complexity of malware threats, research on automated malware detection has become a hot topic in the field of network security. Traditional malware detection techniques require a lot of human intervention and resources (in terms of time and storage) as these involve manual analysis of all malware files in the application. Moreover, malware authors have developed techniques like polymorphism and code obfuscation that evade the traditional signature-based detection approaches employed by anti-virus companies. To solve this issue, malware detection enabled by deep learning (DL) methods is being used recently. This paper presents a performance comparison of 26 state-of-the-art pre-trained convolutional neural network (CNN) models in Android malware detection. It also includes performance obtained by large-scale learning with SVM and RF classifiers and stacking with CNN models. Based on the results, an EfficientNet-B4 CNN-based model is proposed to detect Android malware using image-based malware representations of the Android DEX file. EfficientNet-B4 extracts relevant features from the malware images. These features are then passed through a global average pooling layer and fed into a softmax classifier. The proposed method obtained an accuracy of 95.7% in the binary classification of Android malware images, outperforming the compared models in all performance metrics.