GenIDS Benchmark: Standardized Network Flow Datasets for Cross-Dataset Generalization of Machine Learning-Based Intrusion Detection Systems

GenIDS Benchmark is a collection of standardized network flow datasets developed to support research on the cross-dataset generalization of machine learning-based intrusion detection systems (ML-IDS). The benchmark was constructed from three publicly available intrusion detection datasets: • UNSW-NB15• CIC-IDS2017• CIC-IDS2018 Network flows were extracted from the original PCAP files using NFStream and subsequently processed through a unified preprocessing pipeline to reduce inconsistencies among datasets. The benchmark provides standardized binary and multiclass versions designed for reproducible cross-dataset experiments. These datasets are used in the experiments available in the GenIDS Framework repository, which contains the code for extracting network flow features from PCAP files, preprocessing the datasets, constructing experimental scenarios, implementing the proposed interventions, and training and evaluating the models. The repository also provides the documentation required to reproduce the experiments.