A Standardised Multi-Farm Benchmark Dataset and Evaluation Framework for
Aquaculture Water Quality Prediction, Disease Risk Detection, and Autonomous Sustainability Control
| Metric | Value | Metric | Value |
|---|---|---|---|
| Partner farms | 12 commercial farms | Countries | Egypt, Saudi Arabia, Bangladesh |
| Sensor nodes | 551 IoT nodes | System types | Earthen pond, Biofloc RAS, Polyculture |
| Total readings | 47.8 million (validated) | Disease events | 228 (ground-truth labelled) |
| Species | Tilapia, Shrimp, Catla/Rohu | Split strategy | Chronological 70/15/15 |
72-h water quality forecasting
Primary: DO RMSE (mg/L)
Disease & hypoxia early warning
Primary: DO crash AUROC
5-objective sustainability control
Primary: Pareto Efficiency Score
Cross-farm model transfer
Primary: 72-h DO RMSE
New-farm onboarding speed
Primary: Weeks to target RMSE
Digital twin state estimation
Primary: DT DO RMSE (mg/L)
Pipeline latency benchmark
Primary: E2E latency P99 (ms)
Multi-farm federated learning
Primary: Fleet DO RMSE (mg/L)
| Rank | Model | Site A | Site B | Site C | Fleet Mean | Skill Score |
|---|---|---|---|---|---|---|
| 1 | AquaFarm-X (full) | 0.24 | 0.27 | 0.33 | 0.28 | 0.84 |
| 2 | ST-GCN | 0.44 | 0.36 | 1.01 | 0.60 | 0.66 |
| 3 | Vanilla Transformer | 0.52 | 0.42 | 1.14 | 0.69 | 0.61 |
| 4 | TCN | 0.58 | 0.47 | 1.34 | 0.80 | 0.55 |
| 5 | GRU | 0.64 | 0.52 | 1.46 | 0.87 | 0.51 |
| 6 | LSTM | 0.67 | 0.54 | 1.51 | 0.91 | 0.49 |
| 7 | XGBoost | 0.88 | 0.72 | 1.98 | 1.19 | 0.33 |
| 8 | Persistence | 1.31 | 1.08 | 2.94 | 1.78 | 0.00 |
Submit predictions for any of the 8 tasks.
All submissions are evaluated automatically and retained for reproducibility.