What question did this study set out to answer?

The aim is to enhance network intrusion detection systems by improving their interpretability and robustness against domain shifts.

July 10, 2026Open Access

Explainable and Trustworthy Network Intrusion Detection: From Multi-Class SHAP Analysis to Confidence-Aware Deployment

View Full Paper

Ask AI

Bookmark

Authors

Randomized trial evaluates explainable network intrusion detection in diverse environments, highlighting critical gaps.

The aim is to enhance network intrusion detection systems by improving their interpretability and robustness against domain shifts.
Developed a multi-class IDS pipeline using class-weighted ensemble models and SHAP explainability.
Evaluated on CIC-IDS2017 dataset achieving 98.83% accuracy and macro F1 of 0.826.
Cross-dataset testing on UNSW-NB15 exposed severe domain shifts and overconfidence issues.
Random Forest model achieved 98.83% accuracy but dropped to 31.9% when tested on UNSW-NB15.
Platt scaling improved out-of-domain accuracy to 68.1%, but confidence-based filtering failed to enhance predictions.
XGBoost model retained 98% of samples with high confidence but only had 31.4% accuracy on out-of-domain data.

Vinay Kumar Mandadi (2026) studied this question.

View Full Paper

Ask AI

Bookmark