IMPROVING THE ACCURACY OF CONVOLUTIONAL NEURAL NETWORK BY APPLYING RANDOM SAMPLING METHODS ON NSL-KDD DATASET

Abstract

An Intrusion Detection System (IDS) acts as a cyber security system which monitors and detects any security threats for software and hardware running on the network. Although there have many existing IDS but still face challenges in improving accuracy in detecting security vulnerabilities, not enough methods to reduce the level of alertness and detecting intrusion attacks. Machine learning methods can detect data from past experience and differentiate normal and abnormal data. In this system, the Convolutional Neural Network (CNN) in deep learning method is used for solving the problem of identifying intrusion in a network. NSL – KDD dataset is used to train the data with the CNN algorithm. The system implementation is performed for balanced and unbalanced nature of NSL – KDD dataset. The analysis of evaluation results describes the achievement of the proposed system with the accuracy of 83% in balanced dataset and 80% in unbalanced dataset. The proposed system is implemented by Python programming language on Tensorflow platform.