Explainable Federated Learning for On-Chain UAV Trust Scoring in Blockchain-Based Medical Delivery

The contributions of this work are:

1. A non-IID UAV telemetry generator with four attack classes (GPS spoofing, replay, payload-seal tampering, sensor dropout), filling a gap in public UAV medical
delivery attack datasets.
2. A Byzantine-robust federated trust-scoring pipeline evaluated over a 4 × 4 × 3 grid (cohort × Byzantine ratio × aggregator; 240 runs), cross-validated against
Flower/PyTorch.
3. A 10.3k-parameter Tabular Transformer trust classifier for depot-edge hardware, with a permutation Shapley estimator validated against SHAP/LIME.
4. An atomic on-chain score-plus-explanation-hash audit trail (Hyperledger Fabric)