Convolutional Neural Network based Intelligent Network Intrusion Detection System

Levina Bisen,Sumit Sharma

doi:10.24113/ijoscience.v6i9.317

Abstract

Today cyberspace is developing tremendously, and the Intrusion Detection System (IDS) plays a key role in information security. The IDS, which operates at the network and host levels, should be able to identify various malicious attacks. The job of network-based IDSs is to distinguish between normal and malicious traffic data and trigger an alert in the event of an attack. In addition to traditional signature-based and anomaly-based approaches, many researchers have used various deep learning (DL) techniques to detect intruders, as DL models are capable of automatically extracting salient features from the input data packets. The application of the Convolutional Neural Network (CNN), which is often used to solve research problems in the visual and visual fields, is not much explored for IDS. In this research work the proposed model for intrusion detection is based on feature selection and reduction using CNN and classification using random forest. As compared to some existing work the proposed algorithm proves its efficiency in terms of high accuracy and high detection rate.

Highlights

Intrusion Detection System (IDS) are implemented over host computer or network as a security tool or application to avoid malicious attacks over them
When IDS is applied over multiple systems, connected in a network, it is termed as Network Intrusion Detection System (NIDS)
For analysis this model is designed by applying artificial neural networks (ANNs) in order to detect malicious behavior of coming network traffic

Summary

INTRODUCTION

Intrusion Detection System (IDS) are implemented over host computer or network as a security tool or application to avoid malicious attacks over them. Ryan et al [3] proposed anomaly detection model by using back-propagation 3-layer Multi-Layer Perceptron (MLP) to detect possible attack in network This model analyzed each session logs and analyzed the behavior of data packets. The prediction of coming data packets are analyzed by generalized study of previous known packets For analysis this model is designed by applying artificial neural networks (ANNs) in order to detect malicious behavior of coming network traffic. Similar approach was applied in [5] and [6] using Self-Organizing Maps (SOM) These models are trained on the basis of previously recognized packets and tested over real-time data packets or network traffic. Feng et al [9] integrated SVM and Self-Organized Ant Colony Network for intrusion detection This model is hybrid by merging classification and clustering techniques. DNN model learns the abstract and high-dimensional feature representation of the IDS data

CNN ARCHITECTURE

Post-Processing Phase

RESULT

Findings

CONCLUSION