LIGHTWEIGHT MACHINE LEARNING-BASED INTRUSION DETECTION SYSTEM FOR IOT DEVICES

Abstract

In this study, we compare the capabilities of support vector machines (SVM) and convolutional neural networks (CNN) in network intrusion detection. The classification hypothesis is that incoming network traffic is either benign or malicious and supervised machine learning algorithms are used to train such a classification model. In the current context of increased dependence on digital services and rising cyberattack threats, Intrusion Detection Systems (IDS) have become important tools for the protection of many networks. The main task is the passive monitoring and modification against attacks. It involves the monitoring of network traffic, anomaly detection and blocking of malicious requests. An IDS should be provided training on both normal and anomalous traffic to have efficiency in detecting attacks. We analyze the performance of SVM and CNN in training an IDS to accurately classify network traffic. Feature selection methods were explored, including Correlation-Based, Chi-Square-Based methods, for dimensionality reduction and enhancement of model performance. Our experiments have conclusively demonstrated that the CNN outperforms the SVM in accuracy persistently, making it a good candidate for the IDS.