Hybrid intelligent system for internet of things network attack detection and defense techniques

As Internet of Things (IoT) devices become more integrated into critical infrastructures, they bring vulnerabilities to Denial-of-Service (DoS) attacks, necessitating robust strategies for detection and mitigation. Nowadays, more and more things are connected to the Internet, and the development of new devices is accelerating. The entire communication ecosystem requires security solutions at different levels since these networked smart items can interact with each other in unprotected environments. IoT technology differs from conventional networks in that it has its own set of characteristics, such as different resource limits and network protocol needs. The attacker exploits many security flaws in the IoT system to initiate various attacks. Given the rise in attacks, it is critical to address the implications of the Internet of Things. This paper introduces a novel Hybrid Intelligent System specifically designed to detect and mitigate DoS attacks effectively in IoT environments. Utilizing the CIC-Distributed Denial-of-Service (DDoS) 2019 dataset, the system integrates comprehensive pre-processing techniques, including min-max normalization and an improved SMOTE algorithm, to address the data imbalance. Feature extraction comprises the extraction of raw attributes and statistical measures, standard deviation, median, mean, and variance, Information Gain and Improved Holo-entropy using Spearman's Rank correlation. Classification is performed using a hybrid model that combines Improved LinkNet (ILN) and Long Short-Term Memory (LSTM) architectures, leveraging Cone activation functions to preserve spatial information and enhance training efficiency. Upon detection of attacks, the system identifies and mitigates attacker nodes using threshold-based methods. Practically used to improve security, adjust to new attack techniques, and reduce false alarms in smart cities, smart homes, industrial IoT, and healthcare. The ILN + LSTM Scheme exhibits an accuracy of 0.963, which is superior to the findings of the existing techniques.