Abstract
Hydraulic fracturing (HF) is a pivotal technique in the oil and gas industry, aimed at enhancing hydrocarbon recovery by increasing reservoir permeability through high-pressure fluid injection. Despite its effectiveness, traditional methods used to evaluate HF performance often struggle to capture the complex, nonlinear interactions among operational and geological parameters. This study presents a comprehensive machine learning (ML)-based framework to address this challenge by predicting HF efficiency using three widely used algorithms: Random Forest (RF), Support Vector Machine (SVM), and Neural Networks (NN). The novelty of this research lies in the combined application of advanced statistical characterization and comparative ML modeling over a large-scale dataset comprising 16,000 records. Key statistical metrics, including mean, median, variance, skewness, and quartiles, were used to explore data distribution and inform model training. Additionally, the study uniquely evaluates model robustness across varying train/test data ratios (from 0.1 to 0.9), providing deeper insights into algorithm performance stability. Among the tested models, RF outperformed others by achieving the highest coefficient of determination (R(2) = 0.9804), alongside the lowest Mean Absolute Deviation (MAD) and Root Mean Square Error (RMSE) for both training and testing phases. These results demonstrate RF's capability in handling complex subsurface data with high accuracy and low computational cost. The proposed framework not only enhances predictive accuracy in HF evaluation but also serves as a practical tool for optimizing fracturing design and decision-making in field operations. This integrated approach represents a step forward in applying artificial intelligence for data-driven reservoir engineering and contributes to the advancement of intelligent hydraulic fracturing practices in heterogeneous and data-rich environments.