Vaccine Cold-Chain Cyber-Physical Logistics Dataset

The released dataset, Vaccine Cold-Chain Cyber–Physical Logistics Dataset (VCC-CPLD), comprises 445,603 time-indexed records collected at a 1-minute sampling resolution, covering observations up to 31 December 2025, 23:59:00. Each record represents a cold-chain monitoring snapshot linked to a federated operational site through a client identifier and shipment context attributes such as shipment ID, lot ID, vaccine type, packaging type, route stage, and geographic region code. This structure supports analytical modeling under heterogeneous and non-IID operating conditions distributed across multiple cold-chain logistics nodes.

The feature space integrates multiple cyber–physical dimensions. Environmental telemetry includes internal and ambient temperature, setpoint temperature, rolling minimum/maximum/standard deviation statistics, thermal excursion counts, humidity and condensation indicators, as well as vibration, shock, tilt, and door or light exposure activity. Refrigeration and electronics health signals capture compressor and fan operation, pressure levels, battery and power modes, outage events, defrost and controller resets, firmware versions, and calibration age.

Logistics and mobility descriptors include transport speed, estimated time of arrival, dwell and handoff durations, shipment load status, traffic and weather risk indices, time since pack-out, and cumulative transit duration. Cybersecurity telemetry records network traffic volume, packet transmission rate, failed authentication attempts, scan and DNS anomalies, device join events, TLS failures, and remote session activities. Host integrity logs further document process executions, privilege escalation incidents, file integrity violations, firmware updates, reboot history, CPU/memory/disk utilization, and telemetry log gaps.

Sensor trust indicators quantify telemetry reliability through time-synchronization offsets, sensor ID changes, replay likelihood scores, checksum mismatches, signal jitter, and statistical outlier measures. In parallel, potency and bio-stability proxies include vaccine vial monitor status, cumulative exposure integrals, duration above or below thermal thresholds, freeze events, thermal dose accumulation, stability classification, remaining shelf life estimates, thaw–refreeze counts, and composite potency proxy scores.

For supervised multi-task learning, the dataset provides an attack_type label with six intrusion classes and a continuous potency_remaining_pct_t_plus_H target representing predicted residual potency at a 360-minute forecasting horizon. A derived binary label, safe_to_use_flag_t_plus_H, indicates whether vaccine viability remains within safe administration thresholds at the same prediction interval.