S-ADS: Spectrogram Image-based Anomaly Detection System for IoT networks

Abstract

The Internet of things (IoT) is the smart concept of connecting devices equipped with various sensors, actuators, memory, computational and communicational capabilities using the traditional internet. Since these devices collect a huge amount of sensitive data shared over the internet, the security of an IoT network is of utmost importance due to the more frequent generation of network anomalies. A network-based intrusion detection system (NIDS) is one such tool that can provide much-needed security by shielding the entry points of the IoT network through continuous scanning of network traffic for any suspicious behavior. Recent NIDS experiences low detection accuracy and high false alarm rate (FAR) in detecting network anomalies. To this end, this paper proposes an efficient Spectrogram image-based Anomaly Detection System (S-ADS) using the deep convolutional neural network. The proposed solution is evaluated on the spectrogram images dataset adopted from the Bot-IoT dataset. The experimental results illustrate the effectiveness of the proposed solution by achieving the improvement of 0.4% – 1.9% in the detection accuracy with the reduction in the FAR by 0.6% – 3.5%.