A Novel Approach of Network Intrusion Detection using Deep Learning

Abstract

Protecting computer systems and network against various types of attacks by malicious individual or system has become a vital challenge in recent time. Most or almost all the organizations/agencies maintain their regulatory and day-to-day activities using a connected intranet or internet and hence it becomes vital job to detect security breaches inside the organization’s network. An intelligent network intrusion detection system (NIDS) might play prominent role in surveillance for any possible attacks to the network infrastructures. Various Machine Learning techniques such as Neural Network, SVM, Naïve-Bayes (NB), Random Forests (RF), Decision Tree (DT) and many others already have been used to develop an NIDS. Many of these ML techniques show good prediction accuracy on known attacks, but demonstrate week response on new attacks. Deep Learning (DL) a new branch of Machine Learning has been emerged to show good out-of-sample accuracy since it can automatically learn the features well devising a multi-layered high dimensional space. There are very few works have done in security and in Intrusion detection using Deep Learning. In this work, I have built a novel system leveraging the various Deep Learning techniques such as Deep Autoencoder, Restricted Boltzman Machines and Deep Belief Network for detecting most of the possible attacks in a network infrastructure. This system can also find network attributes which mostly correlate or responsible for a new or old attack with ensuring very good predictive accuracy.