investigates the relationship between training diversity and zero-day detection capability in machine learningbased intrusion detection systems. Rather than develop a new detection algorithm, this research will evaluate how the diversity of attack families present in the training dataset affects the ability of a classifier to detect previously unseen attacks. To accomplish this objective, a diversity controlled evaluation methodology will be developed where classifiers will be trained with various subsets of attack families and then tested on previously unseen families.
Abstract
This repository contains the experimental code used in the paper"Training Diversity and Zero-Day Generalization in Machine Learning-Based Intrusion Detection Systems". The code implements a diversity-controlled evaluation framework for intrusion detection systems where training and testing attack families are separated to simulate zero-day detection scenarios. Experiments are conducted using the publicly available KDD Cup 1999 Intrusion Detection Dataset. Dataset download:https://kdd.ics.uci.edu/databases/kddcup99/kddcup99.html
Fields of Study & Subjects
Network Security and Intrusion Detection
Artificial Immune Systems Applications
Network Packet Processing and Optimization
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 9 March 2026
- Language
- EN