--- license: cc-by-nc-4.0 task_categories: - tabular-classification - tabular-regression - time-series-forecasting language: - en tags: - synthetic - oil-and-gas - midstream - shipping - tanker - logistics - ais - chokepoints - worldscale - bimco - xpertsystems pretty_name: "OIL-031 — Synthetic Shipping & Logistics Dataset (Sample)" size_categories: - 100K distance_nm = haversine(origin, dest) × 0.539957 (km → nm) > maritime_factor = U(1.08, 1.62) (deviation for sea lanes) > chokepoint_count = Poisson(distance / 4500) 7 route regions per actual trade flows: USGC-Asia, MEG-Asia, West Africa- Europe, North Sea-Europe, LatAm-USGC, Med-Europe, Intra-Asia. **`chokepoint_events.csv`** — **7 real EIA chokepoints** with actual traffic shares: | Chokepoint | EIA Traffic Share | Notes | |---|---:|---| | Strait of Hormuz | 21% | Persian Gulf → world (peak risk during Iran tensions) | | Malacca Strait | 16% | Middle East / Africa → Asia (piracy historical) | | Bab el-Mandeb | 12% | Red Sea / Suez (Houthi attacks 2023-2024) | | Suez Canal | 9% | Europe ↔ Asia (Ever Given 2021) | | Cape of Good Hope | 4% | Suez alternative for VLCC (no canal constraint) | | Panama Canal | 3% | Atlantic ↔ Pacific (drought 2023-2024 reduced capacity) | | Turkish Straits | 3% | Black Sea → Mediterranean (Russia oil sanctions 2022) | **`logistics_labels.csv`** — **feature-coupled ML labels**: > delay_risk = clip(total_delay_hours / 120, 0, 1) > congestion_risk = (origin_cong + dest_cong) / 2 > efficiency = 1 - 0.45·delay_risk - 0.25·congestion_risk - 0.15·weather_severity > efficiency_grade = A (≥0.82) / B (≥0.68) / C (≥0.52) / D (<0.52) > recommended_action = reroute (rare_event OR delay > 0.65) > / hold_at_anchor (congestion > 0.7) > / proceed (else) The sample's **delay_risk ↔ efficiency Pearson correlation is r ≈ -0.93** — **near-deterministic inverse coupling validates feature-coupled labels**. --- ## Suggested use cases 1. **Voyage efficiency classification** — 4-class ordinal classifier on `route_efficiency_grade` from delay + congestion + weather features. **Strong feature coupling** — models WILL learn meaningful patterns. 2. **Delay prediction regression** — predict `total_delay_hours` from route + weather + chokepoint + reliability features per delay decomposition formula. 3. **Worldscale freight forecasting** — time-series forecasting of `worldscale_rate` from supply/demand + bunker price + seasonality. 4. **Demurrage cost prediction** — predict `demurrage_usd` from delay hours + charter party type + freight rate features. **Strong physics**: demurrage hours ↔ USD r ≈ +0.83. 5. **Port congestion forecasting** — predict `congestion_index` from `berth_utilization_pct` and queue features per Lloyd's List methodology. 6. **5-class delay type classification** — multi-class classifier on `delay_type` (weather / port / chokepoint / mechanical / rare_event). 7. **Chokepoint risk classification** — 3-class classifier on `risk_level` (low / medium / high) from queue + chokepoint features per EIA chokepoint methodology. 8. **AIS anomaly detection** — anomaly detection on `voyage_events.ais_gap_flag` per IMO Res. A.917 AIS standards. 9. **6-class tanker class classification** — predict `tanker_class` from DWT + capacity + speed features per INTERTANKO classification. 10. **Multi-table relational ML** — entity-resolution + graph neural network learning across the 12 joinable tables via `vessel_id`, `route_id`, `voyage_id`, `cargo_id`, `port_id`. --- ## Loading ```python from datasets import load_dataset ds = load_dataset("xpertsystems/oil031-sample", data_files="logistics_labels.csv") print(ds["train"][0]) ``` Or with pandas: ```python import pandas as pd vessels = pd.read_csv("hf://datasets/xpertsystems/oil031-sample/vessel_master.csv") routes = pd.read_csv("hf://datasets/xpertsystems/oil031-sample/route_master.csv") cargo = pd.read_csv("hf://datasets/xpertsystems/oil031-sample/cargo_movements.csv") events = pd.read_csv("hf://datasets/xpertsystems/oil031-sample/voyage_events.csv") labels = pd.read_csv("hf://datasets/xpertsystems/oil031-sample/logistics_labels.csv") # Multi-table voyage feature engineering: joined = (labels .merge(cargo, on="voyage_id") .merge(vessels, on="vessel_id") .merge(routes, on="route_id")) # Predict route_efficiency_grade from vessel + cargo + route features ``` --- ## Reproducibility All generation is deterministic via the integer `seed` parameter (driving `np.random.default_rng`). A seed sweep across `[42, 7, 123, 2024, 99, 1]` confirms Grade A+ on every seed in this sample. --- ## Honest disclosure of sample-scale limitations This is a **sample** product calibrated for shipping/logistics ML research, not for live voyage planning or chartering decisions. Several notes: 1. **Vessel speed ↔ planned days correlation is moderate (r ≈ -0.18 vs expected -0.5).** Real markets show stronger inverse coupling for fixed distance, but the sample's distance variance dominates the speed-time relationship because routes are randomly sampled across diverse distances. **For speed-time ML, filter to single route_id or single distance bucket** to isolate the speed effect. 2. **Wave height ↔ wind speed correlation is moderate (r ≈ 0.46).** The Beaufort wind scale predicts wave height as a deterministic function of sustained wind, so real-world r is ~0.85-0.95. The sample uses independent N(weather_sev × scale, noise) for both, producing weaker coupling. **For Beaufort-grade ML, derive wave height from wind**: ```python weather['beaufort_wave'] = 0.018 * weather['wind_speed_knots']**2 # ft → m ``` 3. **Season distribution is skewed** (spring 55%, winter 36%, summer 9% for the seed-42 sample). This reflects the 180-day simulation horizon starting January 2024 (covering Jan-Jun → mostly winter/ spring with some summer). **For seasonal-balanced ML, use the full product** (365+ days) or augment with a 4-season cyclic feature. 4. **Tanker class distribution shows mild deviation from declared weights** (sample MR 22.8% vs declared 23%, Aframax 20.4% vs 23%, VLCC 17.6% vs 16%). This is sampling noise at n=250 vessels and converges to declared weights at larger fleet sizes. **For class-balanced ML, use stratified sampling** or filter to specific tanker classes. 5. **Cargo grade distribution is roughly uniform 9-11%** rather than real-world weighted by trade volume. WTI / Brent / Arab Light / Basrah Medium typically dominate VLCC trade (~75% combined per IEA), while Maya / Bonny Light are smaller shares. For realistic trade-flow ML, **filter to specific tanker class × grade combinations** that match real trade routes. 6. **Recommended action is heavily 'proceed' (93%)** because reroute triggers (rare_event OR delay_risk > 0.65) and hold_at_anchor triggers (congestion > 0.7) are rare at sample horizon. For class-balanced recommended_action ML, **oversample rare events** or use the full product (45,000 voyages) for balanced 3-class distributions. 7. **AIS gap flag rate is ~0.6%.** Real AIS coverage is 95-98% globally per UNCTAD, so 0.6% gap rate is realistic for normal operations. But for **AIS-anomaly ML** (detecting sanctions-evasion / "dark fleet" vessels), the sample doesn't generate clusters of correlated AIS gaps. Use the full product or merge with public AIS-spoofing research datasets. 8. **Freight volatility 30d is uniform (mean 22%)** rather than regime-conditioned. Real Worldscale rates have **clustered volatility regimes** per BDTI / BCTI index history (calm periods <15%, volatile periods >40%). For vol-regime ML, **derive your own regime classification** from rolling Worldscale rate statistics. 9. **Charter party type distribution is uniform 24-26% across 4 classes** rather than realistic spot-dominant (~60% spot, ~25% time charter, ~10% COA, ~5% voyage charter per Clarkson commercial reports). For charter-type ML, **filter to specific types** or use derived spot-vs-term classification. --- ## Where physics IS strong (use these for ML) Seven coupling signals in this sample are **physically valid and ML-useful**: | Signal | r | Source | |---|---:|---| | **Delay hours/24 ↔ actual-planned days** | +1.000 | Mass balance of voyage duration | | **Distance ↔ planned transit days** | +0.983 | Kinematics d=v·t per Haversine + speed | | **Delay risk ↔ efficiency score** | -0.932 | Generator's feature-coupled label formula | | **Total delay ↔ efficiency score** | -0.879 | Feature-coupled efficiency formula | | **Demurrage hours ↔ USD** | +0.831 | Commercial demurrage = hours × rate | | **Congestion ↔ queue length** | +0.601 | Port queueing physics (Poisson) | | **Storm severity ↔ weather delay** | +0.534 | Weather delay formula | --- ## Cross-references to other XpertSystems OIL SKUs This SKU is the **first midstream-shipping SKU** in the catalog — opening a new sub-vertical alongside midstream-pipeline (OIL-015/024/025/027) and storage (OIL-028): | Midstream layer | SKU | Focus | |---|---|---| | Pipeline flow assurance | OIL-015 | Wax / hydrate / asphaltene threshold gating | | Pipeline operations | OIL-024 | Hydraulics + SCADA + 15 transient events | | Pipeline leak detection | OIL-025 | Toricelli orifice + acoustic + RBI | | Pipeline corrosion | OIL-027 | de Waard-Milliams + NACE SP0169 CP | | Tank storage | OIL-028 | Mass-balance inventory + API 650/653 | | **Shipping & logistics** | **OIL-031** | **Tanker routes + AIS + Worldscale + chokepoints** *(new sub-vertical)* | **OIL-031 vs OIL-024/025/027**: Pipelines move oil between fixed endpoints. **OIL-031 moves oil across oceans** via 6-class tanker fleet on 7 route regions. Use pipeline SKUs for fixed-asset ML, OIL-031 for **floating- asset chartering / voyage ML**. **OIL-031 vs OIL-028 (storage)**: OIL-028 simulates **stationary tank** inventory dynamics. OIL-031 simulates **moving cargo** dynamics. Natural integration: **OIL-028 + OIL-031** for complete petroleum logistics — storage → vessel loading → voyage → discharge → storage. **OIL-031 + OIL-029 (crude prices)** → freight rates ↔ crude prices for shipping-cost-aware quant trading strategies. **OIL-031 + OIL-030 (supply-demand)** → tanker traffic patterns ↔ regional demand for trade flow ML. --- ## Full product The **full OIL-031 dataset** ships at **5,000 vessels × 12,000 routes × 45,000 voyages × 730 days × 24-hour AIS** (prod mode) producing tens of millions of rows with **realistic trade-flow-weighted cargo grade distributions** (75% WTI/Brent/Arab Light/Basrah on VLCC), **regime- clustered freight volatility** per Baltic Exchange BDTI/BCTI methodology, **realistic spot-dominant charter party mix** (60% spot per Clarkson), **deterministic Beaufort-grade wave-wind coupling**, **AIS-spoofing dark- fleet event generation**, and **chokepoint traffic-share-weighted distribution** per EIA chokepoint data — licensed commercially. Contact XpertSystems.ai for licensing terms. 📧 **pradeep@xpertsystems.ai** 🌐 **https://xpertsystems.ai** --- ## Citation ```bibtex @dataset{xpertsystems_oil031_sample_2026, title = {OIL-031: Synthetic Shipping & Logistics Dataset (Sample)}, author = {XpertSystems.ai}, year = {2026}, url = {https://huggingface.co/datasets/xpertsystems/oil031-sample} } ``` ## Generation details - Sample version : 1.0.0 - Random seed : 42 - Generated : 2026-05-23 13:10:57 UTC - Vessels : 250 - Routes : 500 - Voyages : 2500 - Simulation days : 180 - AIS event step : 24 hours - Tanker classes : 6 (VLCC, Suezmax, Aframax, LR2, MR, Handy) - Real ports : 20 baseline (Houston, Corpus Christi, Rotterdam, Singapore, Fujairah, Ras Tanura, Basrah, Kuwait, Ningbo, Qingdao, Mumbai, Jamnagar, Ulsan, Yokohama, Antwerp, Trieste, Ceyhan, Bonny, Luanda, Santos) - EIA chokepoints : 7 (Suez, Panama, Hormuz, Bab el- Mandeb, Malacca, Turkish Straits, Cape of Good Hope) - Cargo grades : 10 (WTI, Brent, Arab Light, Basrah Medium, Bonny Light, Maya crudes + Diesel, Jet Fuel, Gasoline, Naphtha products) - Delay types : 5 (weather, port_congestion, chokepoint, mechanical_ or_operational, rare_event_disruption) - Charter party : 4 (spot, time_charter, COA, voyage_charter) - Calibration basis : BIMCO, INTERTANKO, Worldscale Association, Baltic Exchange, Clarkson Research, VesselsValue, IMO, MARPOL Annex VI, SOLAS, EIA Chokepoints, UNCTAD, IACS, Lloyd's List Intelligence, AIS per IMO Res. A.917, Beaufort Wind Scale - Overall validation: 100.0/100 — Grade A+