Operational Threshold-Embedded Machine Learning Models for Accurate Hydraulic Fracturing Monitoring
Hydraulic fracturing is critical process in unconventional reservoir development, where accurate real time monitoring directly impacts operational safety, job efficiency and financial aspect. This study presents a hybrid framework integrating domain specific operational thresholds with supervised machine learning for anomaly detection and multi parameter prediction. In anomaly detection, XGBoost achieved the highest accuracy of 0.99 and F1-score of 0.98, outperforming random forest (0.97), logistic regression (0.65), KNN (0.90) across traditional evaluation metrics. In predictive model, gradient boosting algorithm achieved the highest R2 score of 0.97 and the lowest RMSE, outperforming both linear and ridge regression model used for base model. The inclusion of operational constraints into the machine learning model enhanced reliability, prediction accuracy, robustness ensuring practical and actionable results for hydraulic fracturing processes.