AFRAS-FL V2: Reproducibility Artifacts for Exposure-Aware Federated Intrusion Detection Evaluation

This deposit provides reproducibility artifacts supporting the manuscript “Federated Deep Learning for IoT Intrusion Detection: Cross-Device Exact-Feature Recurrence and Training Exposure” by Ali Hussein Fadel, Jaber Karimpour and Zolfaghar Salmanian. The study evaluates FedAvg and FedProx across nine held-out N-BaIoT devices and three fixed random seeds, comprising 54 runs. It distinguishes full-source exact-feature availability from actual sampled-training exposure. Numerical feature equality does not establish packet identity or a causal effect of recurrence. The optimizer comparison establishes neither statistically significant superiority nor equivalence. AFRAS_FL_V2_RAW_REVIEWER_EVIDENCE.zip contains 270 run-level files, comprising 216 unchanged nonreceipt artifacts and 54 redacted receipt derivatives with documented lineage, together with three provenance files and an internal SHA-256 inventory. DEPOSIT_PACKET_MANIFEST.json records the raw archive checksum, size and source commit. AFRAS_FL_V2_REVIEWER_CODE.zip contains the scientific source-code snapshot, protocol and acquisition files, three Stage 19 aggregate-result JSON files and verification instructions. Limitations in reconstructing the historical execution environment are documented in the code archive. Generated research artifacts are licensed under CC BY 4.0. Source code retains its included MIT license. Original third-party data are obtained separately from UCI (https://doi.org/10.24432/C5RC8J) under their applicable terms. Private authorization records, institutional correspondence and author approval records are excluded. This record is published with openly accessible files. The version-specific DOI is https://doi.org/10.5281/zenodo.23777378. File-integrity checks do not constitute independent experimental replication.