Cross-Dataset Performance Drop: Harmonized NSL-KDD and CICIDS2017 Network Intrusion Data

A 5.5 KB Excel file containing harmonized features from two network intrusion datasets, NSL-KDD and CICIDS2017, for evaluating machine learning models. The dataset was created by Shailendra Mishra and last updated on April 20, 2026. It supports a framework for reproducible and statistically validated benchmarking of Intrusion Detection Systems.

Use Cases

• Benchmarking supervised, unsupervised, deep learning, and ensemble models for intrusion detection based on harmonized network traffic features.
• Evaluating the impact of feature harmonization on model performance using ablation analysis as described.
• Conducting statistical validation of model results using Wilcoxon signed-rank, McNemar’s, or DeLong tests on the provided data.
• Studying cross-dataset generalization challenges in machine learning-based security systems.

Strengths

• Features are harmonized across two major intrusion detection datasets (NSL-KDD and CICIDS2017) to enable cross-dataset evaluation.
• The framework is designed for reproducibility and transparency, incorporating a proposed cryptographic logging mechanism.
• Experimental results cited in the description show high performance metrics, such as 98.0% accuracy and 97.0% F1-score for a Random Forest model on the harmonized data.

Limitations

• Row count is unknown, which may limit suitability assessment for large-scale training.
• Column-level documentation is absent; field semantics must be inferred after download.
• The description notes the feature harmonization is based on manual alignment and may not be effective for highly heterogeneous datasets.

Provenance

Source

Features derived from the NSL-KDD and CICIDS2017 network intrusion datasets.

Collection Method

Preprocessing pipeline for feature harmonization, as described in the research framework.

Time Range

null

Freshness

Last updated 2026-04-20 17:36:48; freshness should be verified.

Geography

null