A Random Forest Estimator Combined With N-Artificial Neural Network Classifiers to Optimize Network Intrusion Detection

Information systems have become more complex and highly interconnected. While ensuring real-time connectivity, these systems encounter an increasing amount of malicious traffic. Hence the need to establish a defense method. One of the most common tools for network security is intrusion detection and prevention systems (IDPS). An IDS, while supervising the incoming traffic, tries to identify suspicious activities using either predefined signatures or pre-established user behavior. Signature and behavior based intrusion detection systems are unable to detect new attacks and fall down when facing small behavior deviations. To remedy this problem, many researchers have proposed different approaches for intrusion detection using machine learning techniques as a new and promising tool. Most of the proposed works focus on accuracy over latency and productivity and are tested on the outdated and much criticized kdd99 dataset [1]. In this paper, the authors present a two-level classification framework as a fast, scalable and much accurate traffic classification system, combining early network services identification using a Random Forest estimator and nArtificial Neural Networks for packets classification. The performance of this model is evaluated on the relatively new proposed dataset of New Brunswick University, showing quick classification process with very high accuracy results.