Improvement detection abbility of network attacks by deep learning

Abstract

    The Intrusion Detection System (IDS) is a security software designed to alert automatically when someone or something is trying to infiltrate the system, but this invasion may cause the system to be in danger or violate the privacy policy. Many studies have successfully applied machine learning algorithms to IDS systems that have the ability to self-study and update new attacks. But to limit false alarms and increase the likelihood of predicting attacks, the IDS should have more analytical thinking. This is deep learning. This paper addresses the deep learning as a new approach that can help the IDS system improve accuracy and speed up analysis when input data is too large. With the application of deep neural networks such as the Multilayer Perceptron (MLP) and the Recurrent Neural Network (RNN) on the KDD99 dataset to evaluate Accuracy, Mean Squared Error and Confusion Matrix. The efficiency gains were 98.2% for MLP and 99.04% for RNNs, compared to 92.6% for SVM and 88.46% for Naïve Bayes.