A Single-Domain, Representation-Learning Model for Big Data Classification of Network Intrusion

Classification of network traffic for intrusion detection is a Big Data classification problem. It requires an efficient Machine Learning technique to learn the characteristics of the rapidly changing varieties of traffic in large volume and high velocity so that this knowledge can be applied to a classification task. This paper proposes a supervised-learning technique called the Unit Ring Machine which utilizes the geometric patterns of the network traffic variables to learn the traffic characteristics. It provides a single-domain, representation-learning technique with a class-separate objective for the network intrusion detection. It assigns a large volume of network traffic data to a single unit-ring and categorizes them based on the varieties of network traffic, making it a highly suitable technique for the Big Data classification of network intrusion traffic.