Deep Learning Architecture with Optimal Feature Selec-tion for Intrusion Detection System

Abstract

With the exponentially evolving trends in technology, the networks and computer systems are vulnerable to serious security issues. These technologies are setting the stage for improvements however they are also making it easier for intruders to break into networks without authorization and manipulate the data. Given the fact that the attackers cannot be stopped, their actions can be recognized and avoided by using an Intrusion Detection System (IDS). This paper presents a highly accurate and dynamic ID model in which classification is done using layered network architecture. The primary goal of the suggested work is to improve the intrusion detection rate while minimizing the complexity and dimensionality issues. To accomplish this task, three standard datasets i.e. KDD-CUP99, NSL-KDD and UNSW-NB15 datasets have been utilized in the proposed model with the aim for identifying modern attacks. Fur-thermore, in order to reduce the complexity and dimensionality of the datasets, data cleaning and Feature Selection techniques have been implemented. Finally, the classi-fication is performed by using the multi-layered network model in which data is down sampled and up-sampled with the rise in number of layers for effective training. The usefulness of the system is analyzed and validated in MATLAB software in terms of various performance dependency factors. The outcomes revealed the supremacy of suggested approach in terms of accuracy and False Alarm Rate (FAR) on three da-tasets