Intrusion Detection with Neural Networks: A Tutorial
Alvise De' Faveri Tron · 2022
This chapter builds network intrusion detection systems (NIDS) trained on labeled data which can recognize suspect behavior in a network and classify each connection as normal or anomalous. It provides a complete analysis of the NSL-KDD dataset. The NSL-KDD dataset is provided in two forms: arff files, with binary labels, and csv files, with categorical labels for each instance. The chapter describes all the techniques used for cleaning and preparing the data for the learning phase. Many algorithms are available today for performing feature selection, each with different tradeoffs. Tree-based selection methods use decision trees to derive the impurity of each feature. Univariate feature selection is another common method for feature selection. The chapter provides a comparison between the performances of each neural network model built. It describes which steps should be made to deploy one or more of the produced models in a real network environment.