Scholarlink

<p>Rock mass classification is a prerequisite for geohazard prediction and project construction. Different classification methods and surveyors lead to different rock mass quality classification results, which is a challenge in rock mass quality classification. This paper proposes the CKM-SMOTE method, which accurately evaluates rock mass quality, reduces engineering cost wastage, and prevents disasters. Adopting the machine learning method to evaluate rock mass quality classification results makes up the weaknesses of singleness and subjectiveness for the traditional classification methods and makes the evaluation results more scientific and reasonable. Engineering data sets often contain outliers and unbalanced samples, posing challenges to using machine learning for rock mass quality classification. There are outliers in the evaluation data of rock mass quality, and the number of class I and class V rock masses is too small, which reduces the accuracy of rock mass quality classification. To objectively and accurately obtain rock mass classification results, machine learning methods are introduced to evaluate the rock mass quality. The connectivity-based outlier factor (COF) algorithm is proposed to improve the Kmeans-SMOTE (KM-SMOTE) algorithm (CKM- SMOTE). First, 106 sets of rock mass cases were collected. Next, the KM-SMOTE algorithm was used to address the shortcomings of local expansion and increase the number of rock masses in a few categories, and the COF algorithm was used to eliminate the outliers in the rock mass evaluation data. Finally, 9 machine learning models (the SVM, KNN, NB, DT, MLP, RBF, LDA, XGB, and GBDT models) were used to predict the rock mass grading. The results showed that the accuracy of the 9 algorithms was improved by 31%, 16%, 13%, 17%, 36%, 33%, 15%, 12%, and 17%, respectively, by the CKM-SMOTE algorithm. The model can effectively improve the accuracy of rock mass grading, providing a new method for rock mass quality evaluation. In summary, the CKM-SMOTE algorithm improves the rock mass classification method, solves the problem of imbalance sample data sets, improves the accuracy of the machine learning algorithm in rock mass quality classification, and fills the blank of imbalanced quality classes when applying machine learning algorithm to classify rock mass quality.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Research on the Rock Mass Classification Methods Based on the CKM-SMOTE Algorithm

  • 期刊论文

  • 发表日期：

    2025年6月12日

查看全文
  • Pengneng Chen,
  • Zonghong Zhou,
  • Jian Liu

摘要

Rock mass classification is a prerequisite for geohazard prediction and project construction. Different classification methods and surveyors lead to different rock mass quality classification results, which is a challenge in rock mass quality classification. This paper proposes the CKM-SMOTE method, which accurately evaluates rock mass quality, reduces engineering cost wastage, and prevents disasters. Adopting the machine learning method to evaluate rock mass quality classification results makes up the weaknesses of singleness and subjectiveness for the traditional classification methods and makes the evaluation results more scientific and reasonable. Engineering data sets often contain outliers and unbalanced samples, posing challenges to using machine learning for rock mass quality classification. There are outliers in the evaluation data of rock mass quality, and the number of class I and class V rock masses is too small, which reduces the accuracy of rock mass quality classification. To objectively and accurately obtain rock mass classification results, machine learning methods are introduced to evaluate the rock mass quality. The connectivity-based outlier factor (COF) algorithm is proposed to improve the Kmeans-SMOTE (KM-SMOTE) algorithm (CKM- SMOTE). First, 106 sets of rock mass cases were collected. Next, the KM-SMOTE algorithm was used to address the shortcomings of local expansion and increase the number of rock masses in a few categories, and the COF algorithm was used to eliminate the outliers in the rock mass evaluation data. Finally, 9 machine learning models (the SVM, KNN, NB, DT, MLP, RBF, LDA, XGB, and GBDT models) were used to predict the rock mass grading. The results showed that the accuracy of the 9 algorithms was improved by 31%, 16%, 13%, 17%, 36%, 33%, 15%, 12%, and 17%, respectively, by the CKM-SMOTE algorithm. The model can effectively improve the accuracy of rock mass grading, providing a new method for rock mass quality evaluation. In summary, the CKM-SMOTE algorithm improves the rock mass classification method, solves the problem of imbalance sample data sets, improves the accuracy of the machine learning algorithm in rock mass quality classification, and fills the blank of imbalanced quality classes when applying machine learning algorithm to classify rock mass quality.