Detecting IoT Vulnerabilities: Deep Learning based Intrusion Detection Systems

In this paper, we propose an Intrusion Detection System (IDS) based on Deep Learning (DL), with the use of Convolutional Neural Network (CNN) to detect network intrusions by using the UNSW-NB15 dataset. Our CNN model, that includes two hidden layers of ReLu activation, got accuracy of $\mathbf{8 5}$ percent. However, it performed well at detecting common attack types like “Normal” and “Generic,” and poorly at distinguishing between rare types of attacks like “Worms” and “Backdoor.” On the other hand, compared to the CNN, traditional machine learning (ML) models such as Random Forest, SVM and KNN have more than 96% accuracy. Our findings indicate that while deep learning models have promise, especially for complex and unstructured data, the performance of traditional ML algorithms remains strong for structured network traffic data and is practical for intrusion detection tasks. Hybrid models may also be explored to improve detection of rare attacks in future work.