What question did this study set out to answer?

To evaluate whether explainable artificial intelligence methods can detect classification issues and resolve domain-knowledge discrepancies in anomaly-based network intrusion detection systems.

September 6, 2023

Leveraging Explainable AI Methods Towards Identifying Classification Issues on IDS Datasets

View Full Paper

Ask AI

Bookmark

Authors

ELEric LanferOsnabrück University SSSophia SylvesterNorwegian University of Science and Technology NANils AschenbruckOsnabrück University

Discussion

Member takes

Implication

Experimental study shows that explainable AI identifies dataset imbalances in network intrusion models, suggesting feature explanation tools can diagnose classification errors.

Key Points

To evaluate whether explainable artificial intelligence methods can detect classification issues and resolve domain-knowledge discrepancies in anomaly-based network intrusion detection systems.
Trained Random Forest classifiers on the NSL-KDD benchmark dataset using broad attack classes and generated global feature explanations using SHAP.
Analyzed the internal record distribution of attack classes to identify subcategory data imbalances.
Retrained the Random Forest classifier and SHAP explainer directly on granular attack types to compare model performance and explanation validity.
Initial global explanations from the baseline model deviated substantially from established domain expertise due to severe sample imbalances within attack classes.
Training classifiers directly on granular attack types considerably improved classification performance and generated SHAP explanations aligning with domain knowledge.
XAI methods successfully diagnosed structural dataset biases and overfitting, providing a framework to debug models and improve system trustworthiness.

Cite This Study

Lanfer et al. (2023) studied this question.

synapsesocial.com/papers/6a0f0bda06ecbe833448193a https://doi.org/10.1109/lcn58197.2023.10223401

View Full Paper

Ask AI

Bookmark

Implication

Experimental study shows that explainable AI identifies dataset imbalances in network intrusion models, suggesting feature explanation tools can diagnose classification errors.

Leveraging Explainable AI Methods Towards Identifying Classification Issues on IDS Datasets

Discussion

Member takes

Implication

Key Points

Cite This Study

Leveraging Explainable AI Methods Towards Identifying Classification Issues on IDS Datasets

Discussion

Member takes

Implication

Key Points

Cite This Study

Also Consider

Also Consider