AML-NIDS Survey Data Companion (2022\u20132025)

This repository packages the data companion for \u201cUnderstanding the Adversary: A Survey of Adversarial Machine Learning in Network Intrusion Detection\u201d.Our corpus stems from the PRISMA-based workflow illustrated above: we queried six major digital libraries, deduplicated roughly 8,000 hits down to 4,259 unique records, filtered for venue quality, and manually screened titles, abstracts, and full texts. Progressing through identification, screening, eligibility, and inclusion yielded 94 peer-reviewed primary studies (plus 10 prior surveys) covering AML-for-NIDS research published between 2022 and early 2025.Figure 2 encapsulates the five-axis adversary profile that anchors our survey: each study is characterized by its attack surface, attack phase, adversary position\/access within the network, adversary knowledge assumptions, and adversary goal. Embedding these facets inside the ML-NIDS pipeline clarifies how adversarial leverage propagates from data acquisition through feature extraction to alerting, and provides a consistent frame for the quantitative analyses of RQ1\u2013RQ3.Snapshot of the Taxonomy TableThe unified taxonomy operationalizes the threat-model axes into analyzable layers (metadata, threat-model, methodology, and outcomes), bridging conceptual assumptions with empirical practice across all 94 studies.Paper (short)YearDatasetsModel TypesSurfacePhaseAdversarial attacks against deep learning-based NIDS2022CSE-CIC-IDS-2018CNN; Feed-forward NN; Sequence\/TemporalFeature-levelInference-timeAdversarial attacks against supervised ML-based NIDS2022CIC-IDS-2017Generative; Tree-based; Linear\/ProbabilisticData-\/Feature-levelTraining & InferenceBlack-box attack and NIDS using ML for malicious traffic2022Kitsune-Mirai; CIC-IDS-2017; MAWILab; UNSW-NB15Anomaly-detection; Autoencoder; Feed-forward NN; Generative; Linear\/Probabilistic; Sequence\/TemporalNetwork-levelInference-timeThe complete table is available at data\/raw\/taxonomy_mapping.csv.