A Deep Learning-Based Approach for Intrusion Detection in IoT Networks Using Convolutional Neural Networks

Abstract

As the Internet of Things (IoT) proliferates, the requirement for efficient intrusion detection systems (IDS) becomes increasing. With the increasing rise of Internet of Things (IoT) devices, the need for efficient intrusion detection systems (IDS) has risen too. In this study, a deep learning approach to the IDS system based on Convolutional Neural Network (CNN) on the NSL-KDD dataset is evaluated. A CNN model was built, and we found that the model got 78.27% accuracy on the test set, 98.42% on the training set, and 98.76% on the validation set. A baseline model was built using a Decision Tree classifier which achieved 78% test accuracy. The results show that the new attack type, saint and mscan from the NSL-KDD test set, is difficult, makes the model overfit, and reduces the test performance. However, these issues have not been entirely resolved by dropout regularization and by placing weights on the classes. The shortcomings and vulnerabilities of deep learning in IoT security are explored, and future directions like hybrid learning methods and advanced feature selection are proposed.