EVALUATING SHALLOW AND DEEP NEURAL NETWORKS FOR INTRUSION DETECTION SYSTEMS CYBER SECURITY

Abstract

This project is concerned with intrusion detection systems and several techniques. As we know today’s era is of computer networks or of internet of things which can lead to intrusion and can be devastated to our system, so intrusion detection system can help computer administrators to curb such activities and prevent our systems. As the systems are going towards advancement and in day to day life so as the risk of attack is also going to increase. In this project I am going to discuss about the deep neural networks which is used to calculate the accuracy of the intrusion detection with the learning rate of 0.1 and iteration is 1000. The dataset is used KDD CUP 99 which is a standard set of database which includes large variety of intrusion stimulation in a military network area. We can shield our ICT (Information and communication technology) systems with anomaly detection systems also but they are not that much efficient. They have some fault/foible or we can say demerit such as we might get difficulties/complexity for defining rules of network detection. We have to define each protocol, analyze and implement it and test for the accuracy. Some of the harmful activities that may cause our system might fall in usual usage range which will lead to not recognized through anomaly based that’s why we used IDS which can train and adapt itself after recent novel attacks and becomes indispensable. So in this project I compared results of several classical machine learning techniques like Adaboost , decision tree , KNN , linear regression , Random forest , SVM linear , SVM rbf also used. Deep neural networks with three layers after 100 iteration gives the better results as compared to classical machine learning techniques with the higher accuracy and better results.