An Explainable Hybrid Low-Rank And Deep Learning Framework For Real-Time Iot-Based Image Threat Detection With Robust Zero-Day Attack Classification

Abstract

The number of Internet of Things (IoT) devices is increasing very fast, and because of this, image-based systems are now facing more security problems. These systems are getting affected by different types of cyber-attacks, especially new and unknown attacks called zero-day attacks. The methods we usually use, like machine learning and deep learning, have some problems. They can overfit the data, they are hard to understand, they need more computing power, and they do not work well in real-time when devices have limited resources.To solve these problems, this study presents a new method which is a combination of low-rank approximation and deep learning. The low-rank part helps in finding important patterns in the data by removing extra and unnecessary information. The deep learning part helps in learning complex patterns so that the system can correctly identify different types of threats. By combining both, the model works better and gives more accurate results. For testing, we used a synthetic dataset called Hybrid IoT Image Threat Detection Dataset (HIITD). This dataset includes different types of cases like normal (benign), malware attacks, data theft, and zero-day attacks. The results are very good, and the model achieved 100% accuracy, precision, recall, F1-score, and AUC. Also, the model is very fast. It took only 0.416 seconds for training, and each epoch took around 0.0052 seconds. So, it can be used in real-time systems and edge devices. Another good thing about this method is that it is easy to understand compared to normal deep learning models. It uses structured features, so we can see how the model is making decisions. Overall, this method is simple, efficient, and reliable, and it can be used to improve security in modern IoT systems.