CS Paper Radar

A Three-Axis Stress Test of LLM vs Classical ML for Network Intrusion Detection under Distribution Shift and Adversarial Evasion

Muhammad Ebad Atif, Muhammad Haider Ali

cs.LG cs.AIcs.CR ·submitted 2026-09-11 20:23 UTC · v1

Abstract

Large language models are increasingly benchmarked against classical machine learning for network intrusion detection (NIDS), almost always using same-dataset evaluation, and that protocol turns out to be incomplete. Evaluating XGBoost and RoBERTa-LoRA on two independently collected NetFlow v2 networks across three axes (same-dataset performance, cross-dataset transfer, and adversarial evasion) reveals no universal winner. The two models are statistically tied same-dataset. XGBoost wins decisively under cross-dataset distribution shift, by 15 points of F1 and 25 points of balanced accuracy; on the target network RoBERTa-LoRA's false positive rate reaches 0.78, leaving it barely above chance despite a superficially moderate F1. RoBERTa-LoRA wins decisively under adversarial evasion, by roughly 17 points of F1 at a representative mid-range perturbation strength, while both models hold false positive rates below 0.01 throughout. The model an evaluator would recommend therefore depends entirely on which axis is tested, not on same-dataset accuracy alone. A staged feature-leakage ablation improves cross-dataset transfer non-monotonically, indicating the leakage signal is distributed across the feature representation rather than confined to a few columns, and cross-dataset transfer between our two networks is strongly directional. These results argue for evaluating NIDS models along multiple independent robustness axes, and with more than one metric per axis.

Comment
11 Pages, 5 figures

Closest papers

Nearest neighbours by meaning across the whole index.

Adversarial Robustness Is Not Zero-Day Generalization: A Controlled Evaluation of Flow-Based Intrusion Detectors on Corrected CIC-IDS2017

cs.CR cs.LG ·2026-10-31 ·Abdulhamid Zaro
code/data similarity 0.69
Abstract · 10 pages, 10 tables. Code and per-seed results: https://github.com/abdulhamid-zaro/robustness-vs-zeroday-nids

Synthetic variants of known attacks, including feature-space adversarial perturbations, are sometimes used as stand-ins for unseen ("zero-day") attacks when evaluating machine-learning network intrusion detection systems (NIDS). We test whether this proxy holds. On the corrected CIC-IDS2017 release we train Random Forest, XGBoost, a 1D-CNN and an FT-Transformer and compare, at a fixed 1% false-positive rate over five seeds, detection under constrained perturbations of known attacks (gradient-based and black-box attacks on 42 attacker-controllable features) with leave-one-attack-family-out (LOAO) detection, where an entire family is absent from training. Detection under the strongest perturbation we tested, with budgets of at most 0.2, stays between 90.9% and 99.7%, yet mean detection of held-out families falls to 76.2% (RF), 40.2% (XGBoost), 25.8% (1D-CNN) and 22.8% (FT-Transformer). Robustness exceeds LOAO detection for every model (gaps of 24 to 72 percentage points, 23 to 66 when robustness is also averaged over families; Holm-adjusted p $\le$ 0.00096), and the two measures agree at most on a coarse ranking of the models. An exact mixed-integer evasion search confirms the tree ensembles' scores. Benign-only anomaly detectors and a simple multi-class open-set baseline do not close the gap. A temporal split on 5 seeds (3 of the 4 models) leaves the gap in place. On CSE-CIC-IDS2018 the gap recurs for every model (16 to 68 points). On these benchmarks and under this threat model, a robustness score should not be read as a zero-day score, and the two should be reported separately. Code and per-seed results are publicly available.

0.40

Robustness Evaluation and Detection of Transferable Adversarial Attacks in ML-Based NIDS

cs.CR ·2026-09-05 ·Huda Ali Alatawi
similarity 0.65
Abstract

Machine learning-based network intrusion detection systems (ML-based NIDS) are vulnerable to adversarial evasion, where malicious samples are perturbed to evade detection and be misclassified as benign. Despite growing research on adversarial attacks and defenses for ML-based NIDS, comparative evaluations of multiple attack types, detection models, and defense strategies under a common setting remain limited. In this paper, we evaluate eight adversarial evasion attacks, fifteen detection models, and three representative defense strategies using the NF-UQ-NIDS dataset, which includes recent traditional and IoT network traffic with twenty distinct attack categories. The evaluation compares model performance on clean test data and on robustness evaluation sets that include adversarial samples, analyzes attack success consistency across models, and examines the effect of defense strategies on adversarial robustness. To support model comparison, we introduce the Robustness Index (RI), a compact comparative metric that rewards high balanced accuracy and macro-F1 score computed on the robustness evaluation set while penalizing high attack success rate (ASR). We further present AR-NIDS, a two-stage framework that uses an adversarially trained ensemble to distinguish normal, attack, and adversarial samples, followed by an adversarial attack classifier to identify the attack type. Under the evaluated transfer-based setting, the proposed adversarially trained ensemble achieves the strongest overall trade-off between classification performance on the robustness evaluation set and evasion resistance, reducing the average ASR from 0.41 to 0.03 and achieving an RI of 0.98.

0.08

Provenance, Not Behaviour: A Serialisation Artifact in Edge-IIoTset and a Leakage-Free Benchmark for Precision-Agriculture Intrusion Detection

cs.CR cs.LG ·2026-08-16 ·Mostafa M. Galal
benchmark similarity 0.60
Abstract · 29 pages, 13 tables. Code: https://doi.org/10.5281/zenodo.21941210 Benchmark: https://doi.org/10.5281/zenodo.21941319

Edge-IIoTset is the reference benchmark for machine-learning intrusion detection in the industrial Internet of Things, and results reported on it cluster above 99%. We show that much of that performance is not intrusion detection. The preprocessing recipe distributed with the dataset instructs researchers to one-hot encode seven categorical columns. Four of them separate attack from normal traffic with an accuracy of 1.0000 on their own, through the spelling of the placeholder written for an absent protocol field: the string "0" in the normal-traffic branch of the dataset build against "0.0" in the attack branch. The label is recoverable from a serialisation artifact encoding file provenance, with no network behaviour modelled, and separates every row of both curated subsets. Under 5-fold x 3-repeat cross-validation, five of six standard classifiers attain exactly 1.0000 +/- 0.0000 accuracy and the sixth attains 0.99998. Under a corrected protocol, naive Bayes falls by 0.3005 macro-F1 and the strongest model settles at 0.9503 +/- 0.0011. Label, ordinal and frequency encoding leak identically. Because the curated subsets also lack Modbus and per-device identity, we rebuild the benchmark from the raw captures under uniform parsing, producing AgriEdge: 1,276,122 rows, five devices with full attribution, and no column separating the classes above 0.0288. A leave-one-device-out sweep locates the generalisation boundary at the perception/actuation layer, where random forest falls from 0.9988 to 0.5083 balanced accuracy. Non-IID federated partitioning costs at most 0.0037 macro-F1, but a 20-round LoRaWAN training run costs 4.6 hours of uplink.

0.22

Feature-Context Consistency: Unsupervised Adversarial Detection with Drift Stability on Attributed Graphs

cs.LG cs.AI ·2026-10-12 ·Firas Bayram, Maxime Cordy
top venuecode/data similarity 0.60
Abstract

Adversarial detection on evolving attributed graphs faces two challenges: fraud labels arrive late or not at all, and natural distributional drift erodes any trained decision boundary. Modern detectors can reach strong detection rates when labels are available and the distribution is stationary; sustaining that quality without labels and under drift is the open problem. We identify feature-context consistency -- the alignment between a node's attributes and its aggregated neighborhood attributes -- as a single structural signal that addresses both. We formalize it as the Context Boost (CB) score within a Restricted Boltzmann Machine framework (CB-RBM) and prove two complementary worst-case guarantees: a strict separation between the expected CB of legitimate and adversarial nodes, with gap growing linearly in model capacity, whenever the adversary's context stays far enough from legitimate neighborhood context; and a Wasserstein-Lipschitz bound on how far the CB distribution moves under natural drift. We validate on XBlock Ethereum phishing (~153K nodes) and Reddit banned-user detection (~11K nodes), with two further datasets in the appendix spanning diverse domains. With zero labels, CB-RBM achieves AUROC >= 0.96 on four adversarial attacks and keeps FPR close to its calibrated 5% target under drift -- outperforming supervised GNNs (hundreds of labels), unsupervised graph-anomaly baselines, and a standard RBM whose FPR collapses to 89-100%. What typically requires separate mechanisms -- label-free operation and drift stability -- here follows from a single structural regularity of the data. Code: https://github.com/serval-uni-lu/CB-RBM.

0.47

Multi-Class, Multi-Tier Network Intrusion Detection: A Comprehensive and Reproducible Benchmark

cs.CR cs.LG ·2026-09-28 ·Yufeng Xin, Bryant Goseland, Mohamed Rahouti
benchmarkcode/data similarity 0.59
Abstract

Machine learning (ML) and deep learning (DL) have dominated Intrusion Detection System (IDS) research in recent years. Unfortunately, many existing studies have produced inflated results and unreliable benchmarks due to critical oversights and mistakes in the ML and DL pipeline, from data collection and labeling to feature engineering and model training and evaluation. CIC-IDS2017 is a standard benchmark for network intrusion detection. Still, many published results on this dataset are difficult to compare due to labeling errors, inconsistent flow extraction, potential leakage, and performance evaluation metrics dominated by benign traffic. In this paper, we present a comprehensive benchmark with corrected PCAP-level labeling and a complete evaluation pipeline with diverse ML models. We evaluate eleven tabular classifiers at three nested levels: binary attack detection, nine-class attack-family attribution, and fifteen-class fine-grained classification. A soft-voting ensemble of Random Forest, XGBoost, and LightGBM obtains the best fine-tier macro-F1 of 0.955, with coarse and binary macro-F1 scores of 0.980 and 0.999, respectively. We further conducted a feature selection study based on an analysis of feature importance. This comprehensive benchmark pipeline is configurable and open-source, enabling new feature extraction and model plugins for new datasets. Future work should use this pipeline as a reference point for richer features, rare-class analysis, and model generalization towards new datasets and attack classes.

0.18

What Standard Evaluations Hide About Detectors of Hijacked LLM Agents: Rare Attacks, Unseen Tools and Benchmark Transfer

cs.CR cs.LG ·2026-10-30 ·Mahmoud Abbasi, Juan M. Núñez V., Amin Shahraki
benchmarkcode/data similarity 0.59
Abstract · 14 pages, 4 figures, 6 tables. Code, data and every monitor call: https://github.com/Mahmoud-Abbasi-svg/agent-trace-detector-eval

Detectors can flag hijacked large language model (LLM) agents from the traces of the tools they call. In the eleven evaluations we audited, attacks or anomalies make up 16% to 86% of the test samples, no evaluation varies that share, and six report no test on data from a second benchmark or source. We ask what such evaluations hide. On the AgentDrift and AgentDojo benchmarks we test trained detectors that read tool identities, behavior, call sequences, step embeddings or text, and a released LLM action monitor, AlignmentCheck, with thresholds fixed on training data or set by the monitor's released decision rule. First, at a 2% attack rate a detector with an area under the receiver operating characteristic curve (AUROC) of 0.879 reaches a precision of 0.076, and the mix of negatives alone moves that precision between 0.039 and 0.117. Second, as tool coverage, the share of test tool calls whose tool was seen in training, falls, detector rankings cross near 50% coverage and reverse below it, because unseen tools map to inputs that training never reached. Dropout on tool inputs largely removes the reversal. Third, no detector trained on AgentDrift ranks AgentDojo traces better than chance (AUROC 0.47 to 0.52), including one that reaches 0.996 within AgentDojo. The monitor, trained on neither, reaches 0.930, but with its open backbone its released decision rule would still mostly raise false alerts at a 2% attack rate (precision 0.096). We turn these findings into a checklist for evaluating trace detectors.

0.28

Adaptive and Proximal Federated Optimization for Transferable Network Intrusion Detection

cs.CR ·2026-10-14 ·Rezwan Islam Salvi, MD Rafin Jawad Hafiz, Abeer Mahmood Kalam, Md Moniruzzaman
hot direction similarity 0.59
Abstract · 5 pages, 2 figures

Network Intrusion Detection Systems (NIDS) based on deep learning often fail to detect attack classes that are absent from training data, limiting their robustness against zero-day threats. This property called transferability across attack classes, remains underexplored especially in federated learning (FL) settings where data heterogeneity and client drift pose additional challenges. Prior work (TabFIDS) showed that FedAvg combined with bootstrapping and temporal averaging improves transferability for cross-device NIDS. In this paper, we explore whether other federated optimizers can do better. We compare two optimizers inspired by FedAdam and FedProx against the FedAvg baseline on the CIC-IDS 2017 dataset using an 11-client federated setup, sweep 10 hyperparameter configurations, and contrast two data-balancing strategies: bootstrapping and SMOTE. Using the count of client pairs whose transferred-model accuracy exceeds 70% as our primary metric (following TabFIDS), the proximal optimizer with a strong regularization term yields 48 transferable pairs versus 42 for FedAvg, while the adaptive optimizer with an aggressive server learning rate yields 46 pairs. SMOTE improves per-pair accuracy over bootstrapping but reduces breadth, revealing a depth-vs-breadth trade-off: high accuracy on a few client pairs versus moderate accuracy across most. These results indicate that the choice of federated optimizer, the rule by which client updates are regularized and aggregated into the global model, is very important for federated NIDS.

0.25

Evasion Attacks on Cost-Utility-Based Adversarial Training for Online AutoML in IoT Networks

cs.CR cs.LG ·2026-09-25 ·Chukwunonso Henry Nwokoye, Wajiha Zaheer, Khalil El-Khatib, Li Yang
publishedcode/data similarity 0.59
Abstract · Accepted to the IEEE Middle East Conference on Communications and Networking (MECOM 2026). Code is available at: https://github.com/ChiNonso…

As Internet of Things (IoT) networks increasingly depend on machine learning for anomaly, malware, intrusion detection, and network monitoring, such systems have become attractive targets for evasion attacks. Evasion attacks pose a major security risk because an adversary intentionally modifies input data to mislead a trained model into producing incorrect predictions while evading detection. This study evaluates the impact of black-box evasion attacks on a cost-utility-based adversarial training defense strategy in an Online AutoML context for IoT networks. Specifically, evasion attacks were applied to online learners, including Hoeffding Tree (HT), Leveraging Bagging (LB), Streaming Random Patches (SRP), Hoeffding Adaptive Tree (HAT), and Adaptive Random Forest (ARF). By developing naive and adversarially trained (AT) versions of these online learners, we generated clean and adversarial accuracies for each model. The results show that the AT versions of LB and SRP performed best, achieving the highest adversarial accuracy (0.985) and high clean accuracy (0.993) at the highest cost budget of 1.00, with a maximum accuracy reduction of only 0.8%. Finally, drift detection was conducted using the Early Drift Detection Method (EDDM).

0.42