WilfredthePELE/SPE-DSEATS-Africa-2025-Well-Analysis

# SPE DSEATS Africa 2025 Well Classification Pipeline This repository implements a comprehensive machine learning pipeline for oil and gas well classification and production trend analysis as part of the SPE Africa DSEATS Datathon 2025. The pipeline processes well production data from the provided datasets (spe_africa_dseats_datathon_2025_wells_dataset.csv and reservoir_info.csv), performs data cleaning, feature engineering using TSFresh for time-series features, generates production profiles and visualizations, and trains multi-output classifiers (Random Forest and XGBoost) to predict well attributes such as reservoir name/type, well type, production stability, and trends in GOR (Gas-Oil Ratio), watercut, and productivity index (PI). The model achieves multi-label classification across 7 targets, with hyperparameter tuning via RandomizedSearchCV, stratified cross-validation, and comprehensive evaluation metrics (accuracy, precision, recall, F1-score). Outputs include classified well data, production profile plots, confusion matrices, feature importances, and reservoir oil summaries. Features Data Preprocessing: Cleans numeric columns, handles missing values, assigns wells to reservoirs based on pressure thresholds, and engineers derived metrics (e.g., GOR, watercut, PI, daily production). Time-Series Feature Extraction: Uses TSFresh with EfficientFCParameters for automated feature extraction from production time series. Trend Labeling: Generates ground truth labels for GOR (above/below solution GOR), watercut, and PI trends (Increasing/Decreasing/Flat/Combo). Well Classification: Multi-output Random Forest and XGBoost models predict 7 categorical targets: Reservoir Name/Type, Well Type (GL/NL), Production Type (Steady/Unsteady), and 3 trends. Hyperparameter Tuning: RandomizedSearchCV with stratified K-Fold CV, optimized for F1-weighted score. Visualizations: Individual well production profiles (oil/gas/water, GOR, watercut, PI). Distribution histograms, correlati …