Image-Based IoT Malware Detection Method Using Deep CNNs
Mariam H. Al-Musawi, Ban Mohammed Khammas · 2024
The widespread deployment of Internet of Things (IoT) devices in various fields, such as agriculture and healthcare, has generated a vast amount of data, hence increasing the risk of data breaches. On the other hand, the demand for low-cost and easy-to-use smart products made them vulnerable to malware attacks, which aim to compromise or exploit IoT devices, resulting in damage and data loss. It has increased the necessity for effective and resilient strategies to safeguard these devices from malware threats. The study examines the feasibility of identifying IoT malware with several pre-trained deep convolutional neural network (CNN) models, such as AlexNet, VGG-16, VGG-19, InceptionV3, and MobileNet. The objective of this study is to employ various deep learning models to distinguish between malicious and benign images effectively. This study introduces a relatively new strategy that employs the chi-square goodness-of-fit method on an image dataset prior to model training, aiming to enhance performance and reduce the dataset size since it is noteworthy that this approach is not typically used directly on images. The suggested system's performance has been assessed using the IOT_Malware dataset based on standard performance metrics. The findings of this study indicate that VGG-19 and VGG-16 achieved the highest accuracy among the other models, with scores of 99.09% and 98.55%, respectively, demonstrating their perfect ability to produce the most accurate predictions, followed by MobileNet, InceptionV3, and AlexNet, which attained accuracies of 98.10%, 98.01%, and 97.65%, respectively. Moreover, this research paper presents a state-of-the-art evaluation to add to the system's reliability.