Application of Machine Learning for Effective Screening of Enhanced Oil Recovery Methods
Jawad Ali, Ubedullah Ansari, Fateh Ali, Tariq Javed, Imran Ahmed Hullio
Abstract
Selecting the most suitable enhanced oil recovery (EOR) technique remains challenging due to severe class imbalance in historical datasets and the limitations of traditional screening criteria. To address data imbalance while preserving domain knowledge, this study proposes a novel machine learning framework that incorporates domain-informed synthetic data generation strictly constrained by established EOR screening criteria. An initial dataset of 583 documented EOR projects was compiled from field reports and public databases. After rigorous cleaning, 575 valid samples were retained and subsequently augmented to 760 balanced instances (class sizes ranging from 60–110 samples per class). This reduced the imbalance ratio from 123:1 to approximately 1.8:1. The augmented dataset was processed using principal component analysis (PCA) for dimensionality reduction, followed by hyperparameter tuning and 5-fold cross-validation. Among the evaluated models, K-Nearest Neighbors (KNN) and Random Forest achieved the highest macro-averaged performance (F1-score of 0.89 and 0.85, respectively). The results demonstrate that domain-guided synthetic data generation significantly improves model accuracy and robustness for multi-class EOR screening, offering reservoir engineers a reliable, machine learning-supported decision-making tool.