Abstract
The exponential growth of the Industrial Internet of Things (IIoT) network has increased the risk of malware attacks, which can be a great threat to data integrity and system reliability. Conventional malware detection methods are not always robust, scalable, and interpretable and thus are not effective in very dynamic environments. Accordingly, this paper proposes a meta-ensemble deep learning model that combines VGG16, ResNet50, ResNet101, and DenseNet121 via a meta-learner to classify malware using image-based representations of executable files. The proposed model was evaluated experimentally to have an accuracy of 95%, which was higher in comparison to all of the baseline models, such as VGG 16 (92%), ResNet50 (90%), ResNet101 (89%), and DenseNet121 (88%). The meta-model achieved 97 percent benign precision and 98 percent malware recall in class-wise performance that significantly lowered the false positive rate and increased the detection sensitivity compared to standalone models. The Grad-CAM explainable AI (XAI) method was used to generate visualization of the class-distinctive regions, allowing interpretability and increasing the trust level of the analyst. These findings point to the conclusion that the proposed meta-ensemble framework provides a more scalable, interpretable, and robust solution to malware detection and therefore exhibits promising prospects in the security of IIoT networks.