Intrusion Detection in IoT Environment Using Stochastic Gradient Descent with Warm Restarts and Gated Recurrent Unit
Revatthy Krishnamurthy, Haider Mohmmed Alabdeli, Rajani N, Vankudoth Ramesh, Bura Vijay Kumar · 2024
The rapid growth of Internet of Things (IoT) has attracted the cybercriminals which resulted in increase of cyber-attacks on IoT devices. If the cyber-attacks are undetected for a longer time, it causes serious service interruptions, identity protection threats. To handle these issues, this research proposes a novel intrusion detection system by employing Stochastic Gradient Descent with warm Restarts and Gated Recurrent Unit (SGDR-GRU) method. The data is normalized to obtain common scaled data by using z-score normalization technique and the feature selection is performed by Stochastic Gradient Descent with warm Restarts (SGDR). The SGDR is an optimization technique that escapes local minima, reduces overfitting and improves the performance by adjusting learning rate. Finally, classification of normal and anomaly data is carried out by Gated Recurrent Unit (GRU) that deals with gradient descent problem and captures long-range dependencies. The performance of the proposed model is evaluated on NSL-KDD dataset and achieved high accuracy 99.12%, precision 99.45%, recall 99.23% in identifying intrusion over IoT devices. The proposed method surpassed existing intrusion detection models such as stacked-deep polynomial network (SDPN), Hybrid Chicken Swarm Genetic Algorithm (HCSGA).