Аннотация
The detection of anomalies and the prediction of corrosion defects in oil and gas pipelines constitute critical tasks for ensuring industrial safety and improving operational reliability. This study addresses the problem of regression-based prediction of corrosion defect levels (CR—corrosion defect) using operational process parameters. Machine learning methods, including Decision Tree, Random Forest, LightGBM, and CatBoost, were employed to develop predictive models. Data preprocessing was performed, including feature standardization and hyperparameter tuning using KFold cross-validation. Model performance was primarily evaluated using the Root Mean Square Error (RMSE) on both training and test datasets, as this metric is more sensitive to large prediction errors, which is particularly important in the context of corrosion defect analysis. Additionally, Mean Squared Error (MSE), Mean Absolute Error (MAE), and the coefficient of determination (R2) were used to provide a comprehensive assessment of model accuracy and robustness. Experimental results demonstrate that the CatBoost model achieved the best performance, yielding the lowest RMSE on the test dataset (0.02040) with a close value on the training dataset (0.01682), indicating strong generalization capability. Furthermore, this model outperformed the others in terms of MSE, MAE, and R2 on the test dataset (MSE = 0.000418, MAE = 0.006319, R2 = 0.695544). The obtained results confirm the effectiveness of ensemble methods and gradient boosting algorithms for regression modeling of corrosion defect development processes in pipeline systems.