A Novel Deep Reinforcement Learning (DRL) Method for Detecting Evasion Attack in Iot Environment
Shyamala Ramachandran, D. Prabakaran · 2024
The growing prevalence of Internet of Things (IoT) devices being connected to the internet has resulted in an increase in security concerns, particularly concerning evasion attacks. Reinforcement learning is a category of machine learning in which an agent is taught to make decisions within an environment to maximize a reward signal. In this context, the agent is educated to differentiate network traffic data as either normal or evasive. An evasion attack falls into the category of adversarial attacks, where the attacker seeks to elude detection by exploiting vulnerabilities within the system. This research introduces a novel deep reinforcement learning method for identifying evasion attacks in IoT devices. The proposed design incorporates a neural network and a combination of machine learning classifiers to analyze network traffic data. The features has been extracted using Wireshark, and these features are then utilized as input for a reinforcement learning agent. This agent is trained to distinguish between normal and evasive traffic by combining the outputs of five different classifiers. To train our model, we use two publicly available datasets—one containing benign data and the other containing malicious data. The performance of our approach is evaluated using real-time data, and the results demonstrate that our proposed method surpasses other techniques in terms of accuracy, precision, recall, and F1-score. In conclusion, our deep reinforcement learning approach offers an effective means of detecting and mitigating the risks associated with evasion attacks on IoT devices.