Resrarch On Intrusion Detection Using Manifold Regularization Extreme Learning Machines

With the popularity of Internet, Internet has exposed many problems and this security issues become increasingly prominent. How to detect and identify security problems to ensure the performance, availability and confidentiality of system has become one focus of current researches. Intrusion detection system (IDS) system plays an important role in protecting the safety of network. The intrusion detection model based on machine learning has an advantages of intelligent, automating, and is one research area of intrusion detection technology currently.The Internet is evolving and the existing methods based on machine learning for intrusion detection system requires a lot of labeled data and can’t use unlabeled data. So this paper will bring manifold regularization Extreme learning machines algorithm into the field of intrusion detection, will explore and validate the effectiveness and feasibility of this method, further research on how the number of unlabeled data affect this model and the impact of different kernel functions for this model to provide some guide and direction for the future work. Experimental results show mat this intrusion detection model can use unlabeled data to improve the accuracy of IDS model, the unlabeled data is not the more the better and the sigmoid kernel function or hardlim kernel function is better than the sine, tribas, radbas kernel function.