Network Traffic classification and Abnormal Behavior Detection using Deep Learning

Abstract

Today every developing country is trying to become digital. This digitization makes an increase in usage of the Internet. All business also turned out to be online. Every industry is more over-dependent on the Internet because their business is running on mobile apps, Web application, etc. So business's privacy policy is at risk because of the threat of cybercrimes like identity theft, Denial Of Service (DoS), Phishing Attack, etc. It makes us be attentive to our presence on the Internet. One of the solutions of saving ourselves from being a victim of any cybercrime is Network traffic analysis. Network traffic analysis is the process of classification network packets into two categories, normal and attack. Here We are making a survey of different techniques for network traffic classification and discussed different learning approaches based on normal network traffic behavior of users. We also discuss procedures to detect abnormal behavior of traffic data by Machine Learning (ML) techniques. We are proposing a solution for attack detection using deep learning method. In our proposed solution, we are using a genetic algorithm for feature selection and training a neural network using Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) for sequential time-based data classification. We are using the KDDCUP99 dataset which has 41 features and one target column labeled with a name of the attack for our experiments. KDD99CUP has Four attack categories: DoS, Prob, User to Local (U2L), Remote to User (R2U). We are using 10\% data of KDD99CUP dataset for the experiment.