Deep Learning based Intrusion Detection Technique

Abstract

Intrusion detection system classifies the network traffic to normal and anomaly on the basis of features feeded in classification model. Intrusion detection system monitors and detects unusual traffic on network that have the potential to harm the system by violating security measures. In order to protect personal and sensitive information the demand of intrusion detection system has grown rapidly. In recent years, the use of different type of networks like communication networks and social media networks have been welcomed by people all over the world. Massive growth in network traffic is recorded, however, intrusion threats are not only increased in numbers but also become more sophisticated. Network-based communication has become vulnerable to external and internal attacks. Due to huge traffic and high number of attacks, it is very difficult to inspect incoming traffic which also increase computational cost and time. This fact motivates researchers to develop an intelligent detection system that takes less computational time and provides high accuracy. Different deep learning and machine learning techniques have been applied to handle this problem with better success rate, but there is still work to be done. Deep learning capabilities to detect any malicious activity from complex and heterogeneous traffic make it more favorite for intrusion detection system. In this thesis, we proposed a deep learning-based intrusion detection technique which can detect the intrusion on network data with minimum time. The proposed model is divided into two phases, in the first phase we applied preprocessing and feature selection technique on dataset. In second phase we used Deep Neural Network for classification purpose on preprocessed data. NSL-KDD Binary Class dataset is used for training and testing purpose. The proposed model achieve accuracy of 99.73 using 27 out of 41 features. Features reduction also helps to train and test the proposed model in very less time, which solves the performance and computational time problem for intrusion detection systems.