DEVELOPING ROBUST CYBER SECURITY PROTOCOLS FOR IOT NETWORKS USING AI

Abstract

The exponential expansion of the IoT ecosystem has sparked major cybersecurity worries, despite the fact that it offers unmatched connectivity and ease. The inherent computational limits, extensive distribution, and the heterogeneity of IoT devices are a few of the elements that give rise to these challenges. It is critical to incorporate new technologies into the constantly evolving IoT landscape in order to address these difficulties. Concerning the security of the Internet of Things (IoT), the quickly developing field of machine learning (ML) holds great potential. The widespread use of Internet of Things (IoT) devices has increased cyber dangers in the modern digital age. The Internet of Things (IoT) presents a variety of security risks to these devices, such as encryption, malware, ransomware, and botnets. ttackers can compromise system integrity and demand ransom payments by exploiting and manipulating critical data, which these devices are vulnerable to. Building on lessons learnt from previous cyberattacks, strong cybersecurity protocols are critically important, especially in today's Smart Environments. Our research introduces a new methodology and framework for detecting and countering malware threats in the IoT ecosystem utilising AI approaches in a variety of dispersed contexts. Enhancing security measures in Smart Environments and making them more resilient against future threats, this unique technology proactively monitors network traffic data to detect potential dangers. In order to determine how effective our method was, we ran extensive performance and concurrency tests on the deep neural network (DNN) model that was running on IoT devices. The results showed that there was little effect on power consumption, CPU utilisation, physical memory usage, and network bandwidth, which was very encouraging. When we implemented the DNN model on some IoT gateways, we found that the network bandwidth increased by less than 30 kb/s and that the CPU consumption increased by barely 2%. In addition, the implemented model resulted in an average 13.5% increase in power consumption, even if the memory utilisation for Raspberry Pi devices stayed at 0.2 GB. Not only that, but our ML models showed off some rather impressive detection accuracy rates—roughly 93% accuracy on both datasets and an F1-score of 92%. Our technique successfully identifies dangers in Smart Environments, leading to improved cybersecurity in IoT ecosystems.