An Optimized Multi-Layer Perceptron Framework for Detecting and Classifying IoT Attacks

The Internet of Things (IoT) is revolutionizing industries by connecting billions of smart devices, enabling automation and information exchange. The expansion of IoT ecosystems has simultaneously increased the surface area for cyberattacks. These environments are particularly vulnerable to a wide range of threats, such as Distributed Denial-of-Service (DDoS), poisoning, brute-force SSH intrusions, and various network reconnaissance techniques. The dynamic nature of IoT traffic makes traditional security measures inadequate, thereby necessitating intelligent and adaptive solutions. This study leverages Artificial Intelligence (AI) to combat the growing cybersecurity challenges in IoT. An optimized Multi-Layer Perceptron model is designed to identify and classify cyberattacks with high precision. Using the RT-IoT2022 dataset, which includes realistic network traffic from IoT devices and multiple attack vectors, the model is trained on the 35 most relevant features selected from a total of 85 using permutation importance. The dataset encompasses both benign and adversarial traffic collected via advanced monitoring tools like Wireshark and Zeek. Through rigorous preprocessing, feature engineering, and hyperparameter tuning, the proposed MLP model shows exceptional performance with an accuracy of 99.98. Comparative analysis further shows the superiority of the optimized MLP model over traditional ML algorithms.