Classification method for failure modes of reinforced concrete columns based on imbalanced datasets

Traditional machine learning algorithms have low accuracy to classify the failure modes of minority class due to the imbalanced datasets of failure modes for reinforced concrete columns. In order to address this issue, the initial balanced datasets were generated first by the synthetic minority over-sampling technique (SMOTE). Then the intraclass similarity of datasets was measured based on the weighted K-nearest neighbor algorithm in terms of feature weight. The new balanced datasets for failure modes of reinforced concrete columns were rebuilt by eliminating outliers in the datasets reasonably. A classification method for the failure modes of reinforced concrete columns was developed based on six classical machine learning algorithms and a total of 331 sets of imbalanced data (203 sets of flexure failure, 70 sets of flexure-shear failure and 58 sets of shear failure) for the failure modes of reinforced concrete columns. The results indicate that the accuracy, recall and F1 score of the proposed method for classifying shear failure are increased by an average of 5. 5%, 8. 7% and 7. 2%, respectively, and those for classifying flexure-shear failure are increased by an average of 12. 8%, 15. 7% and 17% respectively, when compared with the traditional methods. With the increase of the unbalance degree of failure mode datasets of reinforced concrete columns, the classification accuracy of the proposed method is improved more significantly. © 2023 Science Press. All rights reserved.