Core code for "Joint evidential-state analysis for unfamiliar-traffic diagnosis in Internet of Things intrusion detection"

Description

Core method implementation for the manuscript "Joint evidential-state analysis for unfamiliar-traffic diagnosis in Internet of Things intrusion detection". This archive contains the evidential neural network, the expected Dirichlet cross-entropy objective with annealed KL regularization, the two-coordinate evidence-state transformation (evidence insufficiency and normalized conditional-allocation entropy), known-reference fitting with the Ledoit-Wolf covariance estimator, and the full and matched-diagonal Mahalanobis scores, together with the model configuration reported in the paper and synthetic unit tests. Scope: this is a core-method release. It does not include raw traffic or packet captures, per-record processed inputs, trained checkpoints, dataset-specific preprocessing and bundle builders, or the plotting and table-production scripts. Source datasets must be obtained from their original providers under their own terms; aggregate numerical summaries supporting the reported results are provided with the manuscript's Supplementary Information. Tested with Python 3.10+; run python -m unittest discover -s tests -v.

Institutions

• Air Force Engineering University

  Shaanxi, Xi'an

Categories

Licence