Аннотация:Security threats for computer networks have increased dramatically over the last decade, becoming bolder and more brazen. There is a strong need for effective Intrusion Detection Systems (IDS) that are designed to interpret intrusion attempts in incoming network traffic intelligently. In this paper, the authors explored the capabilities of Deep Belief Networks (DBN) – one of the most influential deep learning approach – in performing intrusion detection after training with the NSL-KDD dataset. Additionally, they examined the impact of using Extreme Learning Machine (ELM) and Regularized ELM on the same dataset to evaluate the performance against DBN and Support Vector Machine (SVM) approaches. The trained system identifies any type of unknown attack in the dataset examined. In addition to detecting attacks, the proposed system also classifies them into five groups. The implementation with DBN and SVM give a testing accuracy of about 97.5% and 88.33% respectively with 40% of training data selected from the NSL-KDD dataset. On the other hand, the experimental results show around 98.20% and 98.26% testing accuracy respectively for ELM and RELM after reducing the data dimensions from 41 to 9 essential features with 40% training data. ELM and RELM perform better in terms of testing accuracy upon comparison with DBN and SVM.
