Tosa-omokhoa/oil-presence-prediction-spe-dseats-2026

Two-pipeline ML approach (pure data science vs. physics-informed) for predicting oil presence from geological/seismic data — SPE DSEATS Africa Datathon 2026 # Oil Presence Prediction — SPE DSEATS Africa Region Datathon 2026 **Predicting the presence of oil in unexplored locations from geological and seismic properties, using a two-pipeline approach that pairs rigorous data science with petroleum-engineering domain knowledge.** Built by **Team The Cornerstone** for the SPE Data Science and Engineering Analytics Technical Section (DSEATS) Africa Region Datathon 2026. --- ## Team | Name | SPE Number | SPE Section | Affiliation | |---|---|---|---| | Omokhoa Oshose Tosayoname | 5905453 | Port Harcourt | University of Nigeria, Nsukka | | Bakpo Burabari Favour | 5553978 | Port Harcourt | University of Nigeria, Nsukka | | Awor Mboya Celestine | 5425770 | Port Harcourt | Rivers State University | --- ## The Challenge SPE DSEATS Africa Region posed a binary classification problem: given seven geological and seismic properties for a location, predict whether oil is present (`Oil_Presence` = 1) or not (0). - **3,000** training records, **2,000** test records - **72.2% / 27.8%** class split, no oil vs. oil present, a real imbalance problem - Deliberately messy data: nulls, outliers, and technical anomalies the competition guidelines explicitly expect participants to clean using **domain knowledge**, not just imputation defaults | Feature | Description | |---|---| | `Rock_Type` | Sandstone, Limestone, or Shale | | `Porosity` (%) | Empty space in the rock, 0–30% | | `Permeability` (mD) | Ease of fluid movement, 1–1000 mD | | `Trap_Type` | Anticline, Fault, Dome, or None | | `Seismic_Score` | 0–1, likelihood of a reservoir from seismic survey data | | `Proximity_to_Oil_Field` (km) | Distance to nearest known field | | `Estimated_Reservoir_Depth` (m) | 500–5000 m | | `Oil_Presence` | Target: 1 = oil found, 0 = no oil | Full rules: `docs/SPE_DSEATS_Africa_2026_Datathon_Guidelines.pdf` --- ## Our Approach: Two Pipelines, One Question We built **two complete, independently evaluated pipelines** on purpose, not because w …