LAUSR

To address the issues of high computational complexity and insufficient prediction accuracy in traditional wellbore stability prediction methods, this paper employs an approach that integrates rough set theory (RST) with machine learning (ML). First, RST was applied to reduce the attributes of 14 drilling feature parameters that influence wellbore risk levels, eliminating redundant information and retaining the 7 key parameters: displacement, drilling fluid density, azimuth angle, annular pressure loss, RMP, standpipe pressure, and drilling speed. Next, ML techniques were employed to predict wellbore risk levels, with the CatBoost model being selected. This model’s performance was compared with that of random forest (RF) and Extreme Gradient Boosting (XGBoost) models. The results demonstrated that the CatBoost model performed excellently in terms of prediction accuracy, achieving a prediction accuracy of 87.47% without optimization, significantly higher than the RF model (75.69%) and the XGBoost model (72.18%). Furthermore, Bayesian optimization was applied to optimize the hyperparameters of the CatBoost model, resulting in an improved prediction accuracy of 93.25%, which represents a 5.78% increase over the unoptimized version. Finally, the model was applied to predict the wellbore risk levels in different datasets, with the predicted results showing a high degree of consistency with the actual field conditions, thereby demonstrating the model’s practicality and potential.