Anomaly-based intrusion detection system in the Internet of Things using a convolutional neural network and multi-objective enhanced Capuchin Search Algorithm

Abstract

Nowadays, the growth and pervasiveness of Internet of Things (IoT) devices have led to increased attacks by hackers and attackers. On the other hand, using IoT infrastructure in various fields has increased the number of node security breaches, attacks, and anomalies. Therefore, detecting anomalies in IoT devices is vital to reduce attacks and strengthen security. Over the past few years, various research has been conducted in anomaly-based intrusion detection using machine learning and deep learning methods. The biggest challenge in machine learning methods is the inability to extract new features. To do this, researchers use deep learning methods to extract new features that lead to increased accuracy in intrusion detection. There are important unsolved challenges in research, including determining important features in detecting malicious attacks, extracting features from raw network traffic data using deep networks, and insufficient accuracy in detecting attacks against IoT devices. Convolutional neural networks are considered a powerful and reliable method in this field due to the ability to automatically extract features from data and perform faster calculations. This study has designed and implemented the IoT features extraction convolutional neural network called IoTFECNN with hybrid layers for better anomaly detection in the IoT. Moreover, a binary multi-objective enhanced Capuchin Search Algorithm (CSA) called BMECapSA is developed for efficient feature selection. The combination of the IoTFECNN and BMECapSA methods has led to the introduction of a new hybrid method called CNN-BMECapSA-RF. Finally, the proposed method is implemented and tested on two data sets, NSL-KDD and TON-IoT. The results of various experiments exhibit that the proposed method has better results regarding classification criteria compared to existing deep learning and machine learning-based anomaly detection systems. The proposed method has reached 99.99% and 99.85% accuracy by identifying 27% and 44% of the effective features on the TON-IoT and NSL-KDD datasets, respectively.