Lightweight CNN-based malware image classification for resource-constrained applicationsLightweight CNN-based malware image classification for resource-constrained applicationsA. Hota et al.

Malware (Malicious Software) is a malicious piece of code designed with the intention to enter a computer system to carry out harmful operations. It poses a severe threat to computer and internet users. As the number of new malwares increases, the challenge of identifying and classifying them into appropriate families gets more difficult. During the last couple of years, deep convolutional neural network (CNN)-based models resulted promising performance on malware classification. However, the existing deep CNN-based approaches require higher resources in terms of storage and computationally heavy training operations for feeding a large number of data to the CNN model. As a result, the existing approaches are not suitable for malware detection in Internet of things (IoT) applications as IoT-based applications are mostly resource-constrained in nature. To address this issue, exploring lightweight deep learning models for malware detection in resource-constrained devices without compromising accuracy is required. In this regard, the current work proposes a custom lightweight deep convolutional neural network for malware image classification. The proposed technique is used to extract the characteristic features of the malware sample and use the extracted features to classify them into their corresponding malware families by feeding an image to proposed CNN model with Adam optimizer as an input. This furnishes useful information for human analysts without prior knowledge and can be deployed efficiently in IoT applications to aid in developing protection against malwares. The model is validated through multiple trials, demonstrating that the model achieves 96.64% accuracy. The designed CNN model is extremely lightweight and hence can be aptly fit in resource-constrained applications.