Tunnelling and Underground Space Technology· 2026Q1
Intelligent prediction under rockburst class imbalance: A coupled model fusing KMeans-SMOTE and CatBoost
- 0citations
- Q1SCImago
- 2026year
Short summary
A novel KMeans-SMOTE and CatBoost model achieves 82.67% average prediction accuracy and 82.51% Macro F1 score for rockburst prediction, outperforming other ensemble methods on imbalanced data.
AI-generated from the title and abstract; the full text is not read.
Key points
- A KMeans-SMOTE technique was used to balance imbalanced rockburst data, expanding 300 cleaned samples.
- The CatBoost model, trained on balanced data, achieved 82.67% average prediction accuracy and 82.51% Macro F1 score.
- The proposed model outperformed four other ensemble learning algorithms (XGBoost, RF, AdaBoost, LightGBM).
- In an engineering application, the model predicted rockbursts with 95% accuracy on 20 test samples.
AI-generated from the title and abstract; the full text is not read.
Abstract
Solving the problem of data class balance is the key premise to ensure the performance of rockburst intelligent prediction model. This paper adopted an oversampling method (KMeans‑SMOTE) based on K‑means clustering and Synthetic Minority Oversampling Technique (SMOTE). Firstly, this study collected 337 sets of original rockburst samples containing six key features, and obtained 300 sets of valid data after data cleaning. Then, KMeans‑SMOTE oversampling technology was applied to oversample and expand the training data evenly. Subsequently, five ensemble learning algorithms (CatBoost, XGBoost, RF, AdaBoost, LightGBM) were used to train and predict the balanced rockburst dataset. Finally, the model prediction was analyzed using multiple classification evaluation metrics, and the importance of model features was further assessed by combining the SHapley Additive exPlanations interpretability method. The results show that the combined framework of KMeans‑SMOTE and CatBoost achieves good prediction performance, with an average prediction accuracy of 82.67% and a Macro F1 score of 82.51%, outperforming the other four ensemble learning models. In the engineering application at Jiangbian Hydropower Station, the model achieves a prediction accuracy of 95% on 20 test samples, with only one misclassification. This study adopts a prediction strategy for intelligent prediction of rockburst with balanced data.
The authors' abstract, as published at the source. Tunnelling and Underground Space Technology, 2026 · DOI ↗
Continue with a free account
Ask the paper: 3 free questions a day about this paper; save it, get its citation, new summaries every day for your field. Takeaways are Premium.
Continue free on the webSign in with Google or Apple; no card needed. You come back to this paper.
On your phone:
Field: Mechanics of Materials
Mechanics of MaterialsEngineering