Dataset Quality in Machine Learning for Intrusion Detection: An Empirical Benchmarking Study, Design Guidelines, and a Testbed-Validated Pilot

The reliability of machine learning models depends fundamentally on the quality of the datasets used to train and evaluate them, yet dataset quality is often assumed rather than measured. This problem is acute in machine-learning-based intrusion detection, where widely used benchmark datasets suffer from severe class imbalance, inconsistent labelling, data leakage and limited representativeness, pathologies that systematically distort model evaluation. Through a systematic benchmarking study of five datasets across four machine learning and deep learning models, this paper quantifies how these data-quality deficits undermine evaluation: aggregate accuracy diverges from macro-averaged F1 by as much as 0.63, and dataset choice drives more performance variance than model architecture. Building on these findings, this study proposes six quantitative, verifiable design guidelines for constructing high-quality datasets, covering class balance, encrypted-traffic representation, adversarial coverage, multimodal observability, annotation quality and realistic data dynamics. It also operationalises them through a nine-stage data-construction pipeline. The guidelines are validated on a pilot dataset built from a ten-node virtualised testbed, incorporating a per-class deduplication step that exposes how raw sample counts overstate effective data diversity, a data-leakage audit and multi-seed evaluation. The pilot exhibits structural properties: encrypted traffic, eight adversary tactics, and a complete multi-stage attack chain, which are absent from existing benchmarks, demonstrating that the proposed data-quality principles are achievable in practice.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-20
DOI
https://doi.org/10.21203/rs.3.rs-10642513/v1
Primary Topic
Network Security and Intrusion Detection
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Dataset Quality in Machine Learning for Intrusion Detection: An Empirical Benchmarking Study, Design Guidelines, and a Testbed-Validated Pilot

Roohie Naaz Mir, Ifrah Sanober
1 links
Research Square
Network Security and Intrusion Detection
preprint

Dataset Quality in Machine Learning for Intrusion Detection: An Empirical Benchmarking Study, Design Guidelines, and a Testbed-Validated Pilot

Roohie Naaz Mir, Ifrah Sanober
preprint en

Abstract

The reliability of machine learning models depends fundamentally on the quality of the datasets used to train and evaluate them, yet dataset quality is often assumed rather than measured. This problem is acute in machine-learning-based intrusion detection, where widely used benchmark datasets suffer from severe class imbalance, inconsistent labelling, data leakage and limited representativeness, pathologies that systematically distort model evaluation. Through a systematic benchmarking study of five datasets across four machine learning and deep learning models, this paper quantifies how these data-quality deficits undermine evaluation: aggregate accuracy diverges from macro-averaged F1 by as much as 0.63, and dataset choice drives more performance variance than model architecture. Building on these findings, this study proposes six quantitative, verifiable design guidelines for constructing high-quality datasets, covering class balance, encrypted-traffic representation, adversarial coverage, multimodal observability, annotation quality and realistic data dynamics. It also operationalises them through a nine-stage data-construction pipeline. The guidelines are validated on a pilot dataset built from a ten-node virtualised testbed, incorporating a per-class deduplication step that exposes how raw sample counts overstate effective data diversity, a data-leakage audit and multi-seed evaluation. The pilot exhibits structural properties: encrypted traffic, eight adversary tactics, and a complete multi-stage attack chain, which are absent from existing benchmarks, demonstrating that the proposed data-quality principles are achievable in practice.

Research Square
National Institute of Technology Srinagar (IN)
Reduced inequalities
Network Security and Intrusion Detection
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon
Dataset Quality in Machine Learning for Intrusion Detection: An Empirical Benchmarking Study, Design Guidelines, and a Testbed-Validated Pilot — Roohie Naaz Mir, Ifrah Sanober · Research Square (2026) | TGRS Research Map | TGRS