A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
A hybrid cluster-borderline SMOTE method for imbalanced data of rock groutability classification
Abstract Groutability classification is highly important for guaranteeing the safety and quality of grouting projects. However, the precision of groutability classification is inevitably influenced by imbalanced data, in which most fractured rock masses are groutable. Current studies cannot realize high-precision classification for minority classes without considering imbalanced data. Although synthetic minority oversampling technique (SMOTE) is the most influential oversampling method, it produces redundant samples and noise labels. To address these issues, a hybrid cluster-borderline SMOTE method (HCBS) is proposed in this paper. The weights of samples near the minority class center and border were improved to present the category feature, thereby solving the redundant samples problem of SMOTE. To restrain noise, the total samples were divided into different clusters by k-means, and a cluster with an imbalance ratio of more than one was selected to generate new samples. The negative majority samples near the minority class border were removed by variant borderline SMOTE to clean the dataset from noisy instances, and the original and produced minority samples were reduced to a certain ratio according to the majority samples. Finally, the classification precision of the proposed HCBS method was verified using random forest (RF) in grouting engineering, and the number of trees and split variables of the tree nodes were optimized using the grey wolf optimization (GWO) algorithm. The proposed HCBS-GRF method outperformed RF, GWO-optimized RF (GRF), SMOTE-GRF, density SMOTE-GRF, borderline density SMOTE-GRF, and other competitive methods, thereby providing the highest groutability classification accuracy with the least number of generated instances.
A hybrid cluster-borderline SMOTE method for imbalanced data of rock groutability classification
Abstract Groutability classification is highly important for guaranteeing the safety and quality of grouting projects. However, the precision of groutability classification is inevitably influenced by imbalanced data, in which most fractured rock masses are groutable. Current studies cannot realize high-precision classification for minority classes without considering imbalanced data. Although synthetic minority oversampling technique (SMOTE) is the most influential oversampling method, it produces redundant samples and noise labels. To address these issues, a hybrid cluster-borderline SMOTE method (HCBS) is proposed in this paper. The weights of samples near the minority class center and border were improved to present the category feature, thereby solving the redundant samples problem of SMOTE. To restrain noise, the total samples were divided into different clusters by k-means, and a cluster with an imbalance ratio of more than one was selected to generate new samples. The negative majority samples near the minority class border were removed by variant borderline SMOTE to clean the dataset from noisy instances, and the original and produced minority samples were reduced to a certain ratio according to the majority samples. Finally, the classification precision of the proposed HCBS method was verified using random forest (RF) in grouting engineering, and the number of trees and split variables of the tree nodes were optimized using the grey wolf optimization (GWO) algorithm. The proposed HCBS-GRF method outperformed RF, GWO-optimized RF (GRF), SMOTE-GRF, density SMOTE-GRF, borderline density SMOTE-GRF, and other competitive methods, thereby providing the highest groutability classification accuracy with the least number of generated instances.
A hybrid cluster-borderline SMOTE method for imbalanced data of rock groutability classification
Li, Kai (author) / Ren, Bingyu (author) / Guan, Tao (author) / Wang, Jiajun (author) / Yu, Jia (author) / Wang, Kexiang (author) / Huang, Jicun (author)
2021
Article (Journal)
Electronic Resource
English
BKL:
56.00$jBauwesen: Allgemeines
/
38.58
Geomechanik
/
38.58$jGeomechanik
/
56.20
Ingenieurgeologie, Bodenmechanik
/
56.00
Bauwesen: Allgemeines
/
56.20$jIngenieurgeologie$jBodenmechanik
RVK:
ELIB18
Particularities of anti seepage grouting in low-groutability foundation rock
| British Library Conference Proceedings | 2006
Pulverized fly ash - Groutability and effectiveness
| British Library Conference Proceedings | 1996
A new rock-engineering index to assess jointed rock mass groutability
| Taylor & Francis Verlag | 2013
Rock Mass Groutability - Application of Norwegian Experience to Hong Kong
| British Library Conference Proceedings | 2009